ASTROAYODHYA.COM
Admin Panel
Complete AI-Powered Architecture Document
Dual AI Engine: Anthropic Claude (Core Vedic Logic) + Gemini/OpenAI (Content Support)  |  Doc v1.0
Table of Contents
PART 1 — AI ARCHITECTURE
Section 1:  Dual AI Engine — Master Architecture
Section 2:  Anthropic Claude — Core Vedic Logic Engine
Section 3:  Gemini Pro and OpenAI — Supporting Content Engine
Section 4:  AI Routing Rules — Decision Framework
Section 5:  AI Prompt Library — All System Prompts
PART 2 — ADMIN PANEL MODULES
Section 6:  Admin Dashboard — Overview and KPIs
Section 7:  User Management Module
Section 8:  Forecast Management and AI Pipeline
Section 9:  Acharya Management Module
Section 10: Consultation Management
Section 11: Puja Management Module
Section 12: Shop and Product Management
Section 13: Order and Payment Management
Section 14: Membership Management
Section 15: Content and Blog Management
Section 16: Rashifal and Panchang Management
Section 17: SEO and Marketing Tools
Section 18: Foundation Management
Section 19: Notification and Communication Center
Section 20: Analytics and Reporting
Section 21: Financial and Payout Management
Section 22: System Configuration and Settings
Section 23: Queue and Job Monitor
Section 24: Security and Audit Log
Section 25: Admin Roles and Permissions
APPENDIX A: Complete Database Schema
APPENDIX B: API Endpoints Reference
APPENDIX C: Queue Job Reference
APPENDIX D: AI Cost and Token Budget
PART 1 — DUAL AI ARCHITECTURE
SECTION 1 — Dual AI Engine — Master Architecture
1.1  Why Dual AI?
1.2  The Golden Rule
1.3  Master AI Routing Table
SECTION 2 — Anthropic Claude — Core Vedic Logic Engine
2.1  Claude Integration Architecture — Step by Step
Vedic Engine (Python/Swiss Ephemeris) computes full chart JSON — all planetary positions, houses, aspects, Dashas, Shadbala
Chart JSON stored in Redis cache (key: chart:[user_id], TTL: 30 days) — never recomputed unless birth data changes
After forecast order payment confirmed — ForecastGenerationJob dispatched to Laravel Horizon high-priority queue
Worker fetches chart JSON + user language preference + plan type (1yr/2yr/3yr/5yr)
For each chapter (12-19 chapters depending on plan): ClaudeChapterJob dispatched
ClaudeChapterJob calls Claude API via ClaudeForecastService.php with: system_prompt + chart_json + chapter_instructions
Claude returns structured JSON response per chapter — NOT free text
Chapter JSON stored in forecasts.content column (JSON type in MySQL)
After all chapters complete: Acharya review triggered via notification
After Acharya approval: PDFGenerationJob dispatched — Puppeteer renders HTML to PDF
PDF stored in S3 (private bucket) — signed URL generated for user download
User notified via email + WhatsApp + in-app: 'Aapki forecast ready hai'
DEV: Claude API endpoint: https://api.anthropic.com/v1/messages. Use anthropic-php SDK (anthropics/anthropic-sdk-php) via Composer. Store API key in AWS Parameter Store — NEVER in .env file or Git.
2.2  ClaudeForecastService.php — Core Structure
2.3  Claude System Prompt — Master Jyotish Prompt
Master Jyotish System Prompt — key instructions (stored encrypted in DB):
2.4  Chart JSON Input Schema — Sent to Claude
Complete chart JSON structure passed to Claude as user message:
2.5  Claude Output — Structured JSON Response
Claude must return valid JSON (enforced in system prompt). Parsed by Laravel and stored in forecasts.content:
2.6  Claude Token Usage and Cost Control
NOTE: At Rs 99 plan: Claude cost ~Rs 45. Net margin ~Rs 20 + Rs 32 to Foundation. Premium plans have 3x better margins. Monitor via Anthropic Usage Dashboard monthly.
SECTION 3 — Gemini Pro and OpenAI — Supporting Content Engine
3.1  Gemini Pro — Daily Rashifal Generation
Use case: Roz raat 3 AM IST — 12 rashiyon ka daily rashifal generate karna. 9 sections each. 108 content blocks per day.
3.2  OpenAI GPT-4o — Blog Draft Generation
Use case: Admin blog section mein 'Generate Draft' button — GPT-4o se initial draft, editor review karta hai.
3.3  Gemini Flash — Customer Support Auto-Reply
SECTION 4 — AI Routing Rules — Decision Framework
4.1  AI Task Decision Tree
Question 1: Kya is task mein chart data (kundali) se koi interpretation ya recommendation hai?
YES -> Use ANTHROPIC CLAUDE. No exception whatsoever.
NO  -> Proceed to Question 2.
Question 2: Kya yeh task legally or ethically sensitive hai? (Medical prediction, marriage decision, financial guidance)
YES -> Use ANTHROPIC CLAUDE. For careful, nuanced reasoning.
NO  -> Proceed to Question 3.
Question 3: Kya yeh high-volume daily content hai? (Rashifal, social posts, simple support emails)
YES -> Use GEMINI PRO/FLASH. Cost-efficient, fast, excellent Hindi.
NO  -> Proceed to Question 4.
Question 4: Kya yeh English mein creative/structured content hai? (Blog, product descriptions, email copy)
YES -> Use GPT-4o / GPT-4o-mini. Best English creative output.
NO  -> Default: Use GEMINI PRO. Multilingual Hindi tasks.
4.2  Model Selection — Cost vs Quality Matrix
4.3  Fallback Strategy
SECTION 5 — AI Prompt Library
5.1  Claude Prompts Stored in DB
5.2  Gemini Prompts
5.3  OpenAI Prompts
5.4  Prompt Versioning and A/B Testing
PART 2 — ADMIN PANEL MODULES
SECTION 6 — Admin Dashboard — Overview and KPIs
6.1  Main Dashboard — 16 KPI Cards (4x4 Grid)
6.2  Live Activity Feed (Laravel Echo + Pusher)
'New order: 3-Year Forecast — [User Name] — Rs 399'
'Forecast generated: [User] — ready for Acharya review'
'Puja booked: Kaal Sarp Dosha Puja — [User] — Rs 7,099'
'New Premium member: [User] — Rs 399/month'
'Foundation contribution: Rs [X] this month so far'
'AI Alert: Claude API response time exceeded 8 seconds'
'Payment failed: [User] — Rs [X] — retrying in 3 days'
6.3  Revenue and Analytics Charts
SECTION 7 — User Management Module
7.1  User List View
7.2  User Detail Page — 6 Tabs
7.3  User Impersonation (Support Tool)
SECTION 8 — Forecast Management and AI Pipeline
8.1  Forecast Status Pipeline
8.2  AI Generation Controls (Admin Configurable)
8.3  Prompt Editor in Admin Panel
'AI Prompts' section in admin sidebar — Super Admin access only
List all prompts: name | model | last updated | version | active/inactive toggle
Edit view: system prompt textarea + user prompt template + model selector + max tokens + temperature
'Test Prompt' button: paste sample chart JSON -> see live Claude output in real time
'Version History' tab: see all past versions with diff view. Rollback button on each.
'A/B Test' tab: activate two versions simultaneously — compare approval rates
CRITICAL: All prompt changes logged in audit trail with: old content, new content, changed_by, timestamp. Irreversible from UI — requires database restore to undo deleted prompts.
8.4  Acharya Review Interface
8.5  PDF Generation Configuration
SECTION 9 — Acharya Management Module
9.1  Acharya List
9.2  Acharya Onboarding Pipeline — 9 Steps
Acharya submits application via /acharyas/apply form — admin sees in 'Pending Applications' tab
Admin reviews: Brahman lineage certificate + Gurukul certificate + Aadhaar — mark each Verified/Rejected
Phone/video interview — admin records notes in interview_notes field
Demo review assignment — give 1 test forecast to review — evaluate quality and easy Hindi writing
Approve or Reject — if approved: account activated + welcome email + onboarding video sent
Badge assignment: Brahman Verified | Ayodhya Resident | Gurukul Trained | Experience years
Access setup: Acharya admin panel credentials + 2FA setup + mobile app walkthrough
Soft launch: first 5 consultations under supervision — Super Admin monitors
Full activation: profile goes live on /acharyas — available for bookings
9.3  Acharya Availability Management
SECTION 10 — Consultation Management
10.1  Consultation List
10.2  Live Consultation Monitor
'Live Now' tab — all in-progress sessions with: room ID, participants joined, duration elapsed
Connection quality indicator per participant: Green/Yellow/Red
'Join as silent observer' — Super Admin can monitor quality (disclosed in platform terms)
Emergency end session button — for policy violations (misconduct, inappropriate content)
Auto-alert if session runs > 5 minutes past booked end time without extension payment
10.3  Dispute Resolution Workflow
SECTION 11 — Puja Management Module
11.1  Puja Bookings List
11.2  Puja Scheduling Workflow
Booking received -> system auto-assigns to available Acharya matching puja type
If no Acharya auto-available -> 'Needs Assignment' queue -> admin manually assigns
Acharya notified: WhatsApp + email + admin panel notification with sankalp details
Day before puja: automated reminder to Acharya with user's sankalp summary
Puja day: Acharya opens GPS location on mobile -> 'Capture GPS' in their app -> coordinates stored
After puja: Acharya uploads recording -> GPS certificate auto-generated by system
Prasad dispatch: admin marks 'Prasad Dispatched' + enters courier + tracking number
User notified at each step via WhatsApp
11.3  GPS Certificate Generator
Input: booking_id, gps_latitude, gps_longitude, recording_cloudinary_url, acharya_id, puja_timestamp
System generates: PDF certificate with QR code linking to /verify/puja/[certificate_id]
Certificate contains: puja name + yajman name + GPS coordinates + Google Maps link + Acharya signature
Certificate stored in S3 + linked to puja_booking record in DB
User auto-notified: 'Aapka GPS certificate ready hai' — download link in email + in-app
11.4  Puja Catalog Management
SECTION 12 — Shop and Product Management
12.1  Product Management
12.2  AI Product Description Generator
How it works: Admin clicks 'Generate with AI' in product edit form -> enters key facts -> GPT-4o-mini writes Hindi description.
Admin enters: product name + key features + planet association + main benefit
GPT-4o-mini generates: 200-300 word Hindi description + one-line English summary
'Regenerate' button if not satisfied — generates fresh version
Admin edits final text before saving to DB
AI ROUTING: Uses: openai_product_desc prompt | Model: gpt-4o-mini | Cost: approx Rs 0.50 per description
SECTION 13 — Order and Payment Management
13.1  Unified Orders View
13.2  Payment Management
13.3  Razorpay and Stripe Webhook Management
All webhooks logged in webhook_logs table: gateway | event | payload (encrypted) | processed_at | success
Failed webhooks: admin alert + manual reprocess button
Webhook signature verification status shown per record — auto-reject unsigned webhooks
Replay capability: reprocess any past webhook from admin panel
SECTION 14 — Membership Management
14.1  Subscription Dashboard
14.2  Subscription List
14.3  WhatsApp Broadcast Management
SECTION 15 — Content and Blog Management
15.1  Blog Post Management
AI ROUTING: SEO Checker: openai_blog_seo prompt | gpt-4o-mini | Returns JSON: {score, issues[], suggestions[]} | Cost: Rs 0.50/check
15.2  Rashifal Daily Content Management
SECTION 16 — Rashifal and Panchang Management
16.1  Rashifal Generation Control
16.2  Panchang Data Management
SECTION 17 — SEO and Marketing Tools
17.1  SEO Management Panel
17.2  Email Marketing
17.3  Coupon and Discount Management
SECTION 18 — Foundation Management
18.1  Foundation Dashboard
18.2  Foundation Transfer Workflow — Admin Steps
End of quarter: CFO calculates net profit in accounting software
Opens Admin -> Foundation -> New Transfer button
Enters: quarter (Q1/Q2/Q3/Q4), year, total_revenue, operating_costs, net_profit, transfer_amount (50%)
Uploads: bank transfer screenshot as proof
System creates: foundation_transfers record + public impact report entry
'Notify All Users' button: sends email blast + WhatsApp broadcast about transfer
/foundation/impact page auto-updates via ISR — no separate content update needed
Foundation team receives notification to issue acknowledgment letter
SECTION 19 — Notification and Communication Center
19.1  Notification Queue Management
19.2  Support Ticket System
AI ROUTING: Support AI: gemini_support_reply prompt | Agent always reviews — never auto-send | Reduces avg first-response from 4hr to under 30min
SECTION 20 — Analytics and Reporting
20.1  Analytics Dashboard
20.2  Custom Report Builder
SECTION 21 — Financial and Payout Management
21.1  Acharya Payout System
21.2  GST and Tax Compliance
SECTION 22 — System Configuration and Settings
22.1  Platform Settings
22.2  AI Configuration Panel
22.3  Language and Localization
SECTION 23 — Queue and Job Monitor
23.1  Queue Monitor
23.2  Key Job Types and Queue Assignment
23.3  Cron Schedule Reference
SECTION 24 — Security and Audit Log
24.1  Security Configuration
24.2  Audit Log — What Is Recorded
24.3  Data Security Architecture
All PII (birth data, phone numbers, birth locations) encrypted at rest using AES-256
Database credentials stored in AWS Secrets Manager — rotated every 90 days — never in code
AI API keys in AWS Parameter Store — Super Admin access only — rotated quarterly
S3 buckets: private (forecasts, certificates, recordings) + public-CDN (product images only)
HTTPS everywhere — HSTS with 1-year max-age — HTTPS-only cookies
PII excluded from all application logs — only user IDs and order IDs logged
DPDP Act compliance: user data deletion within 30 days of verified deletion request
Database backups: automated daily S3 snapshots — 30-day retention — encrypted
SECTION 25 — Admin Roles and Permissions
25.1  Seven Admin Roles
25.2  Permission Implementation
Using Filament Shield package for granular permissions per resource per role:
NOTE: Each Filament Resource has granular permissions: ViewAny, View, Create, Update, Delete, BulkDelete. Assign per role in admin panel. Changes take effect immediately.
APPENDIX A — Complete Database Schema
A.1  Core Tables Summary
APPENDIX B — Admin API Endpoints Reference
B.1  Internal Admin APIs
APPENDIX C — Queue Job Reference
C.1  All Laravel Horizon Jobs
APPENDIX D — AI Cost and Token Budget
D.1  Monthly AI Cost Projections
NOTE: At 2000 forecasts/day: revenue = Rs 1,56,000/day (avg Rs 78 price). Claude cost = Rs 5,200/day. Cost ratio: 3.3%. Excellent economics. Monitor monthly via Anthropic console.
D.2  Cost Optimization Strategies
Cache chart JSON in Redis 30 days — same chart never recomputed for returning users
Use claude-3-haiku for moderation and safety checks — 60x cheaper than Sonnet
Gemini 1.5 Flash for rashifal — 17x cheaper than Gemini Pro — quality sufficient for daily content
Set hard max_tokens per chapter type — prevents Claude from generating unnecessarily long responses
Batch multiple low-urgency Gemini calls — reduces API call overhead
Monitor top-5 most expensive prompt calls monthly — optimize those first
D.3  Monthly AI Budget Alerts (Configurable in Admin)
AstroAyodhya.com — Admin Panel Document v1.0
STRICTLY CONFIDENTIAL — Internal Developer Reference Only
Dual AI: Anthropic Claude (Core Vedic Logic) + Gemini Pro / GPT-4o (Content Support)
25 Sections + 4 Appendices  |  Filament 3  |  Laravel Horizon  |  Full AI Pipeline
50% profit to larawarefoundation.com  |  A Laraware Group Initiative
— End of AstroAyodhya Complete Documentation Series — 17 Documents — All Pages Covered —
Classification | STRICTLY CONFIDENTIAL — Internal developer document only
Admin URL | admin.astroayodhya.com — subdomain, NOT accessible from main site
Auth | 2FA mandatory. IP whitelist + Laravel Sanctum + TOTP. Role-based access (7 roles).
AI Architecture | Dual AI: Anthropic Claude 3.5 Sonnet (Core Vedic logic) + Gemini Pro / GPT-4o (Content + Support)
Core Principle | All Jyotish computation = Vedic Engine (Python/Swiss Ephemeris). AI only generates TEXT, never computes charts.
Sections | 25 major sections covering every aspect of platform administration
Framework | Laravel 11 + Filament 3 admin UI + Horizon queues + Redis + MySQL + AWS S3
AI Architecture Principle
AstroAyodhya uses a DUAL AI strategy: (1) Anthropic Claude 3.5 Sonnet — for all sacred Jyotish core logic, forecast writing, dosha analysis, and chart interpretation. (2) Google Gemini Pro / OpenAI GPT-4o — for general content, customer support, translations, and non-critical tasks. The Vedic Engine (Python/Swiss Ephemeris) ALWAYS computes chart data — AI only generates human-readable text from that data. AI never calculates planetary positions. Ever.
Claude for Core Vedic Logic | Superior reasoning for complex Jyotish interpretation. 200K context window fits full chart JSON + multi-house analysis in one call. Accuracy critical for trust — no hallucinated numbers. Deeply aligned with classical Parashari reasoning.
Gemini/OpenAI for Content | High-volume lower-stakes tasks: daily rashifal, customer replies, blog drafts, translations. 60-70% cost savings vs using Claude everywhere. Fast, reliable, excellent for Hindi content generation.
Cost optimization | Claude API: approx $15/million input tokens. GPT-4o: approx $5/million. Gemini Pro: approx $1-3/million. Smart routing cuts monthly AI costs by ~65% vs using Claude for everything.
GOLDEN RULE — Never Violate
ANY task involving chart data (kundali) interpretation = ANTHROPIC CLAUDE. ANY generic content task = Gemini or OpenAI. No exceptions. No overrides. This rule is the foundation of user trust.
Task Category | Specific Task | AI Engine | Model
CORE — Kundali | Forecast report writing — all chapters | Anthropic | claude-3-5-sonnet-20241022
CORE — Kundali | Dosha analysis text generation | Anthropic | claude-3-5-sonnet-20241022
CORE — Kundali | Remedy recommendations from chart | Anthropic | claude-3-5-sonnet-20241022
CORE — Kundali | Mahadasha/Antardasha interpretation | Anthropic | claude-3-5-sonnet-20241022
CORE — Kundali | House-by-house life area analysis | Anthropic | claude-3-5-sonnet-20241022
CORE — Kundali | Transit impact on natal chart | Anthropic | claude-3-5-sonnet-20241022
CORE — Tools | Dosha checker explanation text | Anthropic | claude-3-5-sonnet-20241022
CORE — Moderation | Acharya-authored content check | Anthropic | claude-3-haiku-20240307
CONTENT — Rashifal | Daily rashifal — all 12 rashis | Gemini | gemini-1.5-pro
CONTENT — Rashifal | Weekly + monthly rashifal | Gemini | gemini-1.5-pro
CONTENT — Blog | Blog article draft generation | OpenAI | gpt-4o
CONTENT — Blog | SEO title + meta desc + tags | OpenAI | gpt-4o-mini
CONTENT — Support | Customer support auto-reply draft | Gemini | gemini-1.5-flash
CONTENT — Translation | Hindi to English | OpenAI | gpt-4o
CONTENT — Translation | English to Hindi / Regional langs | Gemini | gemini-1.5-pro
CONTENT — Product | Shop product description writing | OpenAI | gpt-4o-mini
CONTENT — Email | Marketing email copy draft | OpenAI | gpt-4o-mini
CONTENT — Social | WhatsApp broadcast / social captions | Gemini | gemini-1.5-flash
MODERATION | User review content moderation | OpenAI | gpt-4o-mini
MODERATION | Push notification copy | OpenAI | gpt-4o-mini
Claude Is the Sacred Engine
Claude handles everything sacred: chart JSON se Jyotish interpretation, dosha analysis, remedy recommendations, Mahadasha guidance. Kisi doosre AI se yeh kaam nahi hoga. Classical Jyotish texts (BPHS, Saravali) ke context mein Claude ka reasoning genuinely superior hai.
ClaudeForecastService.php — Laravel Service Class
use Anthropic\SDK\Anthropic;

class ClaudeForecastService {
  private Anthropic $client;
  private string $model = 'claude-3-5-sonnet-20241022';

  public function generateChapter(
    array  $chartJson,
    string $chapterType,  // career|finance|health|vivah|dasha|etc
    string $planType,     // 1yr|2yr|3yr|5yr
    string $language = 'hi'  // hi|en|hinglish
  ): array {
    $systemPrompt = $this->getSystemPrompt($chapterType, $language);
    $userMessage  = $this->buildUserMessage($chartJson, $chapterType, $planType);

    $response = $this->client->messages()->create([
      'model'      => $this->model,
      'max_tokens' => 4096,
      'system'     => $systemPrompt,
      'messages'   => [['role' => 'user', 'content' => $userMessage]]
    ]);

    $raw = $response->content[0]->text;
    return json_decode($raw, true)  // Claude returns structured JSON
      ?? throw new RuntimeException('Claude did not return valid JSON');
  }
}
System Prompt Security
Yeh system prompt AstroAyodhya ka core IP hai — saalon ki Jyotish knowledge encode ki gayi hai. Store: DB table ai_prompts, encrypted at rest. Never log. Never expose in API response. Admin-only access.
CLAUDE MASTER SYSTEM PROMPT — Vedic Jyotish Engine (abbreviated)
You are a senior Brahman Jyotishacharya with 30+ years Parashari Jyotish practice.
Deep knowledge of BPHS, Saravali, Brihat Jataka. Warm, trusted family-pandit tone.

CRITICAL RULES:
1. NEVER compute planetary positions — use ONLY data in provided chart JSON.
2. Write in Easy Hindi (aam boli) — NOT pure Sanskrit. 'mushkil waqt' not 'pratikulta'.
3. Be SPECIFIC — reference actual planets, houses, dashas from the provided data.
4. NEVER predict death, severe illness, or irreversible negative events.
5. Every difficult period MUST have actionable remedy + positive opportunity.
6. Use 'ho sakta hai' / 'sambhavna hai' — never absolute certainty.
7. NEVER suggest Neelam (Blue Sapphire) without 'consult Acharya first' warning.
8. NEVER suggest 6th/8th/12th house lord gemstone as primary remedy.
9. Output MUST be structured JSON with chapter keys — never free text.
10. Always include: summary, yearly_breakdown[], remedies[], acharya_note.
Chart JSON Schema (Python Vedic Engine Output)
{
  lagna:   { rashi: 'Mesh', rashi_num: 1, degree: 14.5 },
  moon_sign: { rashi: 'Tula', nakshatra: 'Vishakha', pada: 2 },
  planets: {
    Sun:  { house:1, rashi:'Mesh', degree:12.3, nakshatra:'Ashwini',
            status:'exalted', retrograde:false, shadbala:1.42 },
    Moon: { house:7, rashi:'Tula', degree:21.1, nakshatra:'Vishakha',
            status:'normal', retrograde:false, shadbala:0.87 },
    Mars: { house:8, rashi:'Vrishchik', degree:5.7, nakshatra:'Anuradha',
            status:'swagrahi', retrograde:false, shadbala:1.15 },
    // ... all 9 planets same structure
  },
  houses: { 1: { lord:'Mars', planets:['Sun'] }, ... all 12 },
  current_dasha: {
    maha:'Saturn', antar:'Jupiter', pratyantar:'Mercury',
    maha_start:'2023-10-15', maha_end:'2042-10-15',
    antar_start:'2025-01-01', antar_end:'2027-06-01'
  },
  doshas: {
    mangal:   { present:true,  house:8, cancellations:['swagrahi'] },
    kaal_sarp:{ present:false },
    sade_sati:{ present:true,  phase:2, rashi:'Tula' },
    pitra:    { present:false }
  },
  transits: { saturn_rashi:'Kumbh', jupiter_rashi:'Meen', rahu_rashi:'Mesh' },
  plan_type: '3yr',
  forecast_period: { start:'2025-01-01', end:'2028-01-01' },
  language: 'hi'
}
Expected Claude JSON Output — Career Chapter Example
{
  chapter: 'career',
  title: 'Aapka Career aur Kaam-Dhanda — 2025 se 2028',
  summary: 'Shani Mahadasha chal rahi hai. Aapka 10th lord Shani 11th',
           'house mein hai — yeh ek strong position hai business ke liye...',
  yearly_breakdown: [
    { year: 2025,
      theme: 'Nayi shuruwaat ka samay — promotion ya naya project sambhav',
      shubh_mahine: ['March', 'July', 'November'],
      savdhani: ['September mein koi bada financial decision mat lena'] },
    { year: 2026, theme: '...', shubh_mahine: [...], savdhani: [...] },
    { year: 2027, theme: '...', shubh_mahine: [...], savdhani: [...] }
  ],
  remedies: [
    { type:'mantra',   text:'Shani Beej Mantra — roz 108 baar Shaniwar subah' },
    { type:'daan',     text:'Kale til + sarson ka tel Shaniwar ko daan karein' },
    { type:'gemstone', name:'Neelam', note:'PEHLE ACHARYA SE CONSULT KAREIN' }
  ],
  acharya_note: 'Is period mein sabse zaroori hai patience — Shani dhire kaam karta hai...'
}
Plan | Chapters | Avg Tokens per Report | Est. Claude Cost
1-Year (Rs 99) | 12 chapters x 3500 tokens | 42K input + 18K output = 60K total | ~Rs 45 per report
2-Year (Rs 199) | 14 chapters x 4000 | 56K + 22K = 78K total | ~Rs 60 per report
3-Year (Rs 399) | 16 chapters x 4500 | 72K + 28K = 100K total | ~Rs 78 per report
5-Year (Rs 499) | 19 chapters x 5000 | 95K + 38K = 133K total | ~Rs 100 per report
Supporting AI Role
Gemini Pro (Google) aur GPT-4o (OpenAI) handle karte hain: daily rashifal, weekly content, blog drafts, customer support replies, translations, product descriptions, social captions. Yeh tasks Claude se 60-70% sasta hain aur Claude ki unique Jyotish reasoning ki zaroorat nahi.
GeminiRashifalService.php — Daily Rashifal Generation
class GeminiRashifalService {
  public function generateDailyRashifal(string $rashi, array $panchang): array {
    $prompt = $this->buildRashifalPrompt($rashi, $panchang);
    $response = Http::withHeaders(['Content-Type' => 'application/json'])
      ->post(
        'https://generativelanguage.googleapis.com/v1beta/models/'
        . 'gemini-1.5-pro:generateContent?key=' . config('services.gemini.key'),
        [
          'contents' => [['parts' => [['text' => $prompt]]]],
          'generationConfig' => [
            'response_mime_type' => 'application/json',
            'max_output_tokens'  => 2048
          ]
        ]
      );
    $raw = $response->json()['candidates'][0]['content']['parts'][0]['text'];
    return json_decode($raw, true);
  }
}
OpenAIBlogService.php — Blog Draft Generation
class OpenAIBlogService {
  public function generateBlogDraft(string $topic, string $lang = 'hi'): string {
    $client = OpenAI::client(config('services.openai.key'));
    $response = $client->chat()->create([
      'model'    => 'gpt-4o',
      'messages' => [
        ['role' => 'system', 'content' => $this->getBlogSystemPrompt($lang)],
        ['role' => 'user',   'content' => 'Write detailed blog article: ' . $topic]
      ],
      'max_tokens' => 3000,
    ]);
    return $response->choices[0]->message->content;
  }
}
Trigger | New support ticket created via email/WhatsApp
Context sent | Ticket text + user account type + last 3 orders + membership status
Output | Draft reply in Hindi — 3-4 sentences, warm tone, actionable solution
Review | Support agent reviews — edits if needed — THEN sends. AI is assistant not agent.
Never auto-send | Agent ALWAYS reviews before sending. This is mandatory policy.
Model | gemini-1.5-flash — fastest, cheapest. Rs 0.10 per support draft.
Decision Framework
Yeh decision tree follow karo jab bhi koi naya AI task aaye. Is tree ke bahar mat jaao.
Model | Cost (approx per M tokens) | Best For | Context Window
claude-3-5-sonnet-20241022 | $15 input / $75 output | Core Vedic logic, all kundali tasks | 200K tokens
claude-3-haiku-20240307 | $0.25 input / $1.25 output | Moderation, quick checks, safety | 200K tokens
gpt-4o | $5 input / $15 output | Blog drafts, product copy, English content | 128K tokens
gpt-4o-mini | $0.15 input / $0.60 output | SEO tags, push notifications, summaries | 128K tokens
gemini-1.5-pro | $1.25 input / $5 output | Rashifal, Hindi content, translations | 2M tokens!
gemini-1.5-flash | $0.075 input / $0.30 output | Support drafts, social captions, quick tasks | 1M tokens
Claude fails/timeout | Retry 3x with 5s delay. If still fails: FAILED status + admin alert + manual Acharya write.
Gemini fails | Fallback to GPT-4o-mini for same task. Log the failure.
OpenAI fails | Fallback to Gemini for same task. Log the failure.
All AI fails — rashifal | Use yesterday's content with modified date intro. Alert admin immediately.
All AI fails — forecast | Mark FAILED. Acharya writes manually. User refunded if >2 hours delay.
Cost spike alert | Daily spend > Rs 5,000 — auto-email to Super Admin. Review and optimize.
Prompt Security
All prompts stored in DB table: ai_prompts. Encrypted at rest (AES-256). Editable ONLY by Super Admin from admin panel. Full version history maintained — roll back anytime. Never logged in plain text. Never in Git.
Prompt Name | Chapter/Use | Key Instructions | Ver
claude_forecast_career | Career chapter | 10th/6th house analysis, Dasha lord on profession, year-wise shubh mahine | v1.2
claude_forecast_finance | Finance chapter | 2nd/11th house, wealth yogas, investment timing, avoid periods | v1.1
claude_forecast_health | Health chapter | 6th/1st house, Moon strength, vulnerable seasons, precautions | v1.1
claude_forecast_vivah | Marriage chapter | 7th house, Navamsa D9, Venus, Dasha timing, Mangal Dosha impact | v1.3
claude_forecast_dasha | Dasha chapter | Full Maha-Antar timeline, best + challenging sub-periods | v1.0
claude_forecast_spirituality | Spirituality chapter | 12th/9th house, Ketu position, dharma path | v1.0
claude_dosha_mangal | Mangal Dosha check | All 25 cancellations, severity, marriage + health impact, remedies | v2.0
claude_dosha_kaalsarp | Kaal Sarp check | 12 types identification, partial vs complete, life impact areas | v1.1
claude_dosha_sade_sati | Sade Sati check | 3 phase analysis, per-rashi effects, year-wise guidance | v1.0
claude_remedy_gemstone | Gemstone recommendation | 14 rules including 6/8/12 lord filter, combustion check, retrograde note | v1.5
claude_remedy_rudraksha | Rudraksha recommendation | 27-nakshatra mapping, planet weakness check, wearing instructions | v1.0
Prompt Name | Use | Language | Model
gemini_rashifal_daily | Daily rashifal for all 12 rashis — 9 sections each | Hindi primary | gemini-1.5-pro
gemini_rashifal_weekly | Weekly rashifal with transits | Hindi | gemini-1.5-pro
gemini_rashifal_monthly | Monthly rashifal — comprehensive | Hindi | gemini-1.5-pro
gemini_support_reply | Customer support draft reply | Hindi + English | gemini-1.5-flash
gemini_translate_hi_en | Hindi to English article translation | Both | gemini-1.5-pro
gemini_translate_en_hi | English to Hindi translation | Both | gemini-1.5-pro
gemini_social_caption | WhatsApp broadcast + social captions | Hindi | gemini-1.5-flash
gemini_product_desc | Product descriptions for shop items | Hindi | gemini-1.5-pro
Prompt Name | Use | Notes | Model
openai_blog_draft | Blog article draft — Hindi or English | 1500-2500 word output | gpt-4o
openai_blog_seo | SEO title + meta desc + tags + schema | Returns structured JSON | gpt-4o-mini
openai_email_marketing | Marketing email copy | Persuasive, easy Hindi, CTA focused | gpt-4o-mini
openai_review_moderate | User review content moderation | Returns: approve/flag/reject + reason | gpt-4o-mini
openai_push_notification | Push notification copy | Under 100 chars, engaging Hindi | gpt-4o-mini
Version history | Every prompt change stored: old content + new content + changed_by + changed_at
Rollback | One-click rollback to any previous version from admin panel
A/B testing | Run two prompt versions simultaneously — compare Acharya approval rates and report quality scores
Test prompt | Admin can enter sample chart JSON and see live Claude output before going live
Prompt analytics | Per-prompt: avg tokens used, avg response time, failure rate, cost per call
Framework: Filament 3
Admin panel built on Laravel Filament 3. URL: admin.astroayodhya.com. All modules are Filament Resources and Pages. Authentication: Filament Auth + mandatory 2FA (TOTP via Google Authenticator). IP allowlist enforced.
Row 1 — Revenue | Row 2 — Users | Row 3 — Operations | Row 4 — AI Health
Today Revenue (Rs) | New Signups Today | Forecasts In Queue | Claude API Calls Today
MRR (Monthly Recurring) | Active Members | Pujas Pending | Gemini API Calls Today
Forecast Revenue MTD | DAU (Daily Active Users) | Acharya Reviews Pending | OpenAI API Calls Today
Foundation Transfer QTD | Total Registered Users | Support Tickets Open | Total AI Cost Today (Rs)
Daily revenue chart | Bar chart — last 30 days — breakdown by: Forecasts, Pujas, Shop, Membership, Consultations
MRR growth line | Last 12 months membership revenue growth
Product mix pie | Forecasts % | Pujas % | Shop % | Membership % | Consultations %
Foundation tracker | Quarterly transfers with targets vs actuals — green/red bar
User growth | Cumulative users + daily new — last 6 months — two-line chart
AI cost trend | Daily AI spend by model — Claude vs Gemini vs OpenAI — stacked bar
Columns displayed | ID | Name | Email | Phone | Rashi | Membership Plan | Orders Count | Joined Date | Last Login | Status
Quick filters | By: Membership plan | Rashi | Registration date range | Has kundali | Active/Inactive | Country
Search | Instant search: name, email, phone number — debounced 300ms
Bulk actions | Export CSV | Send bulk notification | Suspend multiple | Delete (soft)
Per-user actions | View full profile | Edit data | Impersonate (support) | Send direct message | View all orders
Profile tab | All personal data + kundali info + edit capability. Birth details with impact warning if changed.
Orders tab | All orders: forecasts, pujas, shop, consultations — sorted newest first
Membership tab | Current plan + billing history + UPI mandate status + WhatsApp connection status
Kundali tab | All saved charts — view D1 inline + planet table + dosha flags + Dasha timeline
Foundation tab | Estimated lifetime contribution to Foundation from all purchases
Audit tab | Complete activity log: logins, purchases, data changes, support tickets
Security Policy
Impersonation (acting as a user for support) MUST be logged in audit trail: who impersonated, which user, at what time, from which IP. ONLY Super Admin and Support Manager roles can impersonate. Auto-expires in 15 minutes. Red banner visible to admin during impersonation: 'You are viewing as [User Name]'.
Most Complex Module
Forecast management is the most AI-intensive module. Manages the complete pipeline: kundali compute -> Claude write -> Acharya review -> PDF generate -> deliver. Every step monitored and alertable.
COMPUTING | Vedic Engine (Python) computing chart JSON — typically < 3 seconds
WRITING | Claude generating chapters — 5-15 minutes for full report
ACHARYA_REVIEW | Waiting for Acharya to review and approve — SLA: 2 hours max
FORMATTING | Puppeteer generating PDF from approved content — typically < 2 minutes
DELIVERED | PDF ready in S3 — user notified — download link active
FAILED | Any step failed — admin alerted — manual intervention needed
Setting | Description | Default | Allowed Range
Claude model | Which model for forecast generation | claude-3-5-sonnet-20241022 | claude-3-5-sonnet or claude-3-haiku
Max tokens/chapter | Token limit per chapter generation | 4096 | 1000 to 8000
Temperature | Claude creativity (0=precise, 1=creative) | 0.7 | 0.5 to 0.9
Retry attempts | Times to retry failed Claude calls | 3 | 1 to 5
Chapter sequence | Order of chapter generation | Career,Finance,Health... | Admin reorderable
Auto-approve hours | Auto-approve if no Acharya action | 2 hours | 1 to 8 hours
Parallel chapters | Chapters generated simultaneously | 3 at once | 1 to 5
Review queue | All forecasts assigned to this Acharya — oldest first. SLA timer visible.
Review view | Full forecast rendered in beautiful HTML — chapter by chapter with PDF styling
Action bar | Approve | Edit Section (inline) | Add Personal Note | Request Regeneration | Flag Issue
Edit mode | Rich text editor per chapter — Acharya rewrites in easy Hindi
Personal note | 300-char max — Acharya's personal observation — appears in PDF
Reference panel | Right sidebar: user's D1 chart + planet table + current Dasha — for reference while reviewing
Timer | Soft timer showing review duration. Target: 10 min per report. Not enforced.
Template editor | Admin edits PDF HTML/CSS template. Uses Tailwind CSS. Live preview.
Cover page | User name + order ID + forecast period + AstroAyodhya logo + saffron design
Chapter headers | Each chapter with custom icon + saffron heading + chapter summary
Acharya section | Acharya photo + name + verification badge + personal note
Remedies section | Formatted remedy cards — mantra, daan, gemstone, puja recommendations
Footer | Page numbers + 'Prepared for [Name]' watermark + AstroAyodhya branding
Foundation note | Last page: Foundation mission + how this report helped Foundation
Test PDF | 'Generate Test PDF' button — uses dummy chart data — preview before going live
Columns | ID | Name | Photo | Status | Specializations | Avg Rating | Total Reviews | Consultations Done | Reports Reviewed | Monthly Earnings
Status badges | ACTIVE | ON_LEAVE | SUSPENDED | PENDING_VERIFICATION | TRAINEE
Filters | By: Status | Specialization tag | Rating range | Ayodhya-based | Consultation-enabled
Actions | View profile | Edit | Suspend | Payout history | Review queue | Availability calendar
Weekly schedule | Acharya sets available days + time slots via their admin panel
Admin override | Force-block any slot for training, meetings, platform events
Bulk schedule | Set Acharya availability for entire week in one action — batch update
Calendar view | Color-coded: Green = available, Red = booked, Yellow = Acharya-blocked, Grey = unavailable
Capacity | Admin sets max consultations per day per Acharya — prevents burnout
Leave management | Acharya marks leave in advance — bookings blocked for those dates automatically
Columns | ID | User | Acharya | Date/Time | Duration | Status | Amount | Has Recording | Review Given
Status | BOOKED | CONFIRMED | IN_PROGRESS | COMPLETED | CANCELLED | NO_SHOW | DISPUTED
Filters | By: Acharya | Status | Date range | Duration | Amount | Has recording
Actions | View full details | Issue refund | Resolve dispute | Join as silent observer (Super Admin)
Dispute raised | User or Acharya raises dispute from their dashboard — reason + description
Admin notification | Immediate alert to Admin Manager + support team
Evidence review | Recording URL + chat log + both party statements visible in dispute panel
AI assist | GPT-4o-mini summarizes consultation chat log for admin — flags key moments
Resolution options | Full refund | Partial refund (50%) | No refund | Reschedule free | Warning to Acharya
Timeline | Resolution required within 48 hours of dispute raised. Auto-escalate to Super Admin after 24hr.
Columns | ID | User | Puja Name | Tier | Assigned Acharya | Scheduled Date/Time | Status | Amount | Samagri Ordered | GPS Captured
Status flow | BOOKED -> CONFIRMED -> SCHEDULED -> IN_PROGRESS -> COMPLETED -> CANCELLED
Actions | View details | Assign Acharya | Reschedule | Cancel+Refund | Mark GPS Captured | View Certificate | Track Prasad
Puja list | All 25+ pujas — name, slug, price tiers, puja model, active status, booking count
Edit puja | All fields: name (hi+en), description, pricing tiers, samagri list, what user receives, SEO
Add new puja | Full creation form — goes live after admin activates toggle
Availability toggle | Active/Inactive — inactive pujas hidden from /puja page immediately via ISR
Seasonal blocking | Block puja bookings during Chaturmas (Aug-Oct) — automated or manual
Product list | SKU | Name | Category | Price | Sale Price | Stock | Status | Last 30d Sales | Rating
Add product | Form: name hi+en | category | price | sale_price | images (multi-upload) | planet | mukhi (for rudraksha) | description | caution_level | consult_required
Edit product | All fields editable + SEO fields (meta title, description) + schema markup
Stock management | Update count. Low stock threshold configurable per product. Auto-alert at threshold.
Images | Multiple upload with drag-drop reorder. Auto-compress to WebP. CDN via Cloudinary.
Energization | Upload Acharya energization video + assign Acharya who performed it + GPS timestamp
QR generation | One-click QR code generation — unique per product unit — stored in DB + printed on package
Order types | Forecasts | Shop products | Pujas | Consultations — all in one list with type badge
Columns | Order ID | Type | User | Amount | Payment Status | Order Status | Created At | Actions
Filters | By: type | payment status | order status | date range | amount range | user
Export | CSV export with active filters applied — for monthly accounting reconciliation
Payment list | All transactions: Razorpay (INR) + Stripe (USD) — with gateway status
Failed payments tab | Dedicated view — manual retry button + 'Send payment link' to user
Refund tool | Issue partial or full refund — amount editable — reason required — syncs to gateway
Daily reconciliation | Automated: DB orders vs Razorpay/Stripe dashboard — mismatch alert to Finance Manager
GST reports | Monthly summary: B2C invoices + B2B (GSTIN users) — GSTR-1 compatible CSV export
Webhook logs | All incoming webhooks: event_type + payload + processing_status + timestamp
Active members | Count by plan: Basic | Premium | VIP. MRR per plan. Total MRR.
Churn metrics | Monthly churn % — last 3 months trend. Churn reason breakdown.
Trial conversions | % of 7-day trials converting to paid — by plan type
Dunning status | Failed payment recovery: how many in grace period, how many converted
Columns | User | Plan | Status | Next Billing Date | Amount | WhatsApp Connected | Paused Until | Started
Status | TRIAL | ACTIVE | PAUSED | CANCELLED | FAILED | PENDING
Actions | View | Edit plan | Pause | Cancel | Apply discount coupon | Manually retry payment
Composer | Write message + select audience: all members | Basic only | Premium+ | specific rashi
Scheduling | Queue message for specific date/time — e.g. daily rashifal at 7AM IST
Delivery stats | Per broadcast: Sent count | Delivered | Read (where available) | STOP replies
Template management | WhatsApp message templates — view Meta approval status for each
Opt-outs | Auto-maintained list of users who replied STOP — never contacted again
AI draft | 'Draft with Gemini' button — generates broadcast text — admin edits then sends
Post list | All posts: title | author | category | status | published_at | views (30d) | SEO score
Statuses | DRAFT | IN_REVIEW | PUBLISHED | SCHEDULED | ARCHIVED
New post editor | Tiptap rich text editor with Hindi Devanagari support + image upload + SEO fields
AI draft button | 'Generate with AI' -> enter title + category -> GPT-4o writes initial Hindi draft
SEO checker | 'Check SEO' -> AI analyzes heading structure, keyword density, meta length -> scores out of 100
Auto-fix | 'AI Optimize' button: AI rewrites title/meta for better SEO — admin reviews before saving
Schedule | Set future date/time for automatic publication — useful for festival content
Daily rashifal view | All 12 rashis x 9 sections grid — shows today's generated content for review
Inline edit | Click any cell (rashi x section) — opens inline editor with Hindi text
Regenerate section | 'Redo with Gemini' on any single section — fresh content in 3 seconds
Bulk approve | 'Approve All for Today' — marks entire day's rashifal as reviewed and ready
Pre-generate | System pre-generates 3 days ahead — admin can review in advance
Quality flag | Flag any section as 'Needs Review' — prevents it from going out in WhatsApp
Cron log | Generation history — last 30 days. Success/fail per run. Execution time.
Manual trigger | 'Generate Today's Rashifal' button — runs Gemini immediately without waiting for cron
Preview mode | See tomorrow's pre-generated content before it goes live
Prompt tuning | Direct link to gemini_rashifal_daily prompt — edit quality parameters
ISR trigger | After bulk edit: 'Revalidate Next.js' button — pushes changes to live site instantly
Data source | Swiss Ephemeris Python engine — auto-calculated daily for all 50 cities
City management | Add/edit/remove cities from Panchang city list — each city with lat/lng/timezone
Daily validation | Automated sanity check: if any city data seems anomalous — admin alert
Manual override | Correct specific Panchang value if calculation error detected — with reason field
Festival calendar | Manage festival entries shown in Panchang: name (hi+en), date, significance, tithi
Vivah muhurat | Annual vivah muhurat list: date, time, details — admin adds/edits/removes
Sitemap | Auto-generated sitemap.xml — 'Regenerate' button after major content changes
Robots.txt | Direct editor — update crawling rules without code deployment
Meta defaults | Site-wide og:image, Twitter card defaults, structured data base settings
301 Redirects | Add/edit/delete URL redirects — for old content URLs, rebranded pages
Core Web Vitals | Google Search Console data embedded in admin — monitor all green/red signals
Hreflang | Auto-generated for multilingual pages — review table + override per URL
Campaign list | All campaigns: name | status | audience size | sent count | open rate | click rate
Composer | Subject + HTML body + audience segment + send time + A/B subject test option
AI draft | 'Draft with GPT' -> enter goal -> gpt-4o-mini writes persuasive Hindi email body
Segments | Pre-built: All users | Members only | Non-members | Last 30d inactive | By rashi | By purchase type
Transactional logs | All transactional emails: user + type + sent_at + opened + clicked — searchable
Deliverability | Postmark reputation score + bounce rate + spam rate — visible in dashboard
Coupon list | Code | Type (% or fixed) | Discount | Used count | Max uses | Expiry | Status
Create coupon | Percentage OR fixed amount | specific plan/product/category | expiry date | max uses limit
Referral codes | User-specific referral codes — track conversions + reward referrer
Auto-apply rules | E.g.: first purchase = 10% off, 6-month member = 5% shop discount — rule engine
Revenue tracker | Running quarterly revenue total + auto-calculated 50% Foundation transfer amount
Transfer log | All past Foundation transfers: quarter | amount (Rs) | bank proof URL | acknowledgment URL | date
Impact numbers | Update Foundation program stats shown across entire site — one-click update all pages via ISR
Quarterly report | 'Generate Report' — pulls stats + AI formats intro copy (GPT-4o-mini) + admin edits + publishes
Automation Win
Transfer record in admin = Foundation impact page at /foundation/impact auto-updates instantly. Users see real-time transparency. No manual CMS update needed.
Queue view | All pending notifications: channel (WhatsApp/Email/Push/SMS) | user | scheduled_at | status
Manual blast | Compose + audience selection + channel + schedule time + preview before send
Template library | All notification templates — edit copy, placeholders, CTA links per template
Delivery reporting | Per notification: sent | delivered | opened | clicked | failed — download CSV
Frequency controls | User-level frequency caps — e.g. max 3 WhatsApp messages per day per user
Ticket list | All tickets: source (WhatsApp/Email/In-app) | user | subject | status | assigned_to | SLA timer
Status flow | OPEN -> IN_PROGRESS -> PENDING_USER_REPLY -> RESOLVED -> CLOSED
AI first response | Gemini Flash generates draft reply within 30 seconds of ticket creation — agent reviews + sends
Auto-assignment | By category: Technical -> Dev team | Billing -> Finance | Puja -> Ops India
SLA alerts | Red highlight: ticket > 24hr open. Orange: > 8hr. Slack/email alert to team lead.
WhatsApp integration | Reply to WhatsApp tickets directly from admin — real-time bidirectional
Resolution analytics | Avg resolution time by category | agent performance | satisfaction scores
Traffic analytics | Daily/weekly/monthly visitors + sessions + bounce rate + top pages
Conversion funnel | Tool visit -> Email signup -> Forecast purchase — drop-off rates at each stage
Revenue analytics | By product type | by time period | by user cohort | by acquisition channel
User behavior | Most-used tools | most-read blog articles | shop bestsellers | avg session time
AI cost analytics | Per-model daily spend: Claude vs Gemini vs OpenAI — trend + budget remaining
WhatsApp analytics | Delivery rate | read rate | STOP rate | per-broadcast comparison
Retention cohorts | User retention by signup month — see which cohorts stay longest
Revenue report | Date range + product filter -> revenue by category + payment method + GST breakdown + CSV
User report | New users by source channel | retention by cohort | active vs churn | geographic split
Acharya performance | Per Acharya: reviews/month, consultations/month, earnings, rating trend, response time
Foundation report | Quarterly transfer details + program allocation + impact numbers — formatted for public
AI cost report | Per-model cost breakdown + token usage + cost per report/rashifal/blog — weekly/monthly
SEO report | Organic traffic + ranking keywords + Core Web Vitals — pulls from Search Console API
Payout list | Acharya name | period | reports_reviewed x rate | consultations x 70% | total due | status
Calculation | Auto: (reports_reviewed * per_report_rate) + (consultation_fees * 0.70) — generated weekly
Approve payout | 'Process Payout' -> verify UPI ID -> initiate transfer -> mark PAID + receipt number
Payout rates | Forecast review: Rs 15-25/report. Consultation: 70% of booking fee. Bonus for volume.
Payout history | Full statement per Acharya — downloadable PDF + CSV for their records
Failed payouts | Alert if transfer fails — retry mechanism + alternative payment method option
Monthly GST | B2C invoice summary + B2B (GSTIN users) — GSTR-1 compatible CSV export for CA
TDS (26Q) | Acharya payouts > Rs 30,000/year — TDS deduction tracking + Form 16A generation
Annual summary | Total revenue | GST collected | Foundation transfer | Acharya payouts — CA package
Invoice generation | Auto-generate GST invoices for all orders — downloadable by users from /dashboard
Export formats | CSV + Excel + PDF — all financial reports — date range filterable
General | Site name, tagline, contact emails, support WhatsApp, default timezone (IST)
Feature flags | Toggle on/off: Consultations | Shop | Membership | Blog | Tools | any individual feature
Maintenance mode | Enable: site shows maintenance page. Specific IPs (dev team) stay accessible.
API keys storage | All secrets in AWS Parameter Store — Razorpay | Stripe | Twilio | Meta WABA | Anthropic | Gemini | OpenAI | Google Maps
Email config | Postmark SMTP settings + from address + reply-to + email footer template
Storage config | AWS S3 bucket names + Cloudinary credentials + CDN base URLs
Model defaults | Global default model per task type — override per individual task if needed
Token budgets | Max tokens per task type — hard ceiling to prevent runaway costs
Rate limits | Max AI API calls per hour per model — protects against abuse or misconfiguration
Cost alerts | Email alert thresholds: daily > Rs 5,000 | weekly > Rs 25,000 | monthly > Rs 1,00,000
AI logs | Every AI call: model | prompt_name | input_tokens | output_tokens | cost | response_time | status
Prompt library | All prompts with version history — see Section 5 for full spec
Active languages | Toggle each of 20 languages active/inactive — inactive = no content shown in that lang
Default language | Hindi — not editable. Platform-level decision.
UI strings | Key-value translation editor for UI text in all active languages
Hreflang config | Auto-generated hreflang tags review table — override per-URL if needed
Language priority | Hindi > English > Hinglish > Regional — fallback chain when content missing
Horizon Dashboard
Laravel Horizon at admin.astroayodhya.com/horizon — real-time queue monitoring. Super Admin access only. Shows all queues, workers, jobs, throughput, failures.
Pending jobs | Count per queue: high-priority | default | low-priority. Backlog alert if > 100.
Processing | Currently running jobs: name + started_at + estimated completion + worker ID
Failed jobs | All failures: job name + error message + stack trace + retry button + delete
Throughput chart | Jobs processed per minute — last 60 minutes — detect slowdowns
Worker scaling | Number of active workers per queue — restart all workers button
Job Name | Queue | Priority | Triggered By
ForecastGenerationJob | high | Critical — SLA 30 min | Order payment confirmed webhook
ClaudeChapterJob | high | Critical — per chapter | ForecastGenerationJob dispatches
GeminiRashifalJob | high | Daily at 3AM | Cron scheduler
PanchangGenerationJob | high | Daily at 2AM | Cron scheduler
PDFGenerationJob | default | Normal | After Acharya approval
EmailDeliveryJob | low | Normal — async | Any event triggering email
WhatsAppBroadcastJob | default | Normal | Daily 7AM cron
WebhookProcessJob | high | Urgent — < 5s | Razorpay/Stripe webhook received
ISRRevalidationJob | default | Normal | After content publish/update
FoundationReportJob | low | Quarterly | After Foundation transfer recorded
AcharyaPayoutCalcJob | low | Weekly Friday 9AM | Cron scheduler
UserActivityTrackJob | low | Background | Various user events
0 2 * * * | 2 AM IST — Panchang generation for all 50 cities
0 3 * * * | 3 AM IST — Daily Rashifal generation via Gemini (12 rashis x 9 sections)
30 3 * * * | 3:30 AM IST — Next.js ISR revalidation for rashifal + panchang pages
0 3 1 * * | 3 AM on 1st — Monthly rashifal generation
0 7 * * * | 7 AM IST — WhatsApp daily rashifal broadcast to all members
*/5 * * * * | Every 5 minutes — Acharya availability status refresh to Redis
0 */6 * * * | Every 6 hours — Foundation stats counter cache refresh
0 0 * * 1 | Monday midnight — Weekly payment reconciliation with Razorpay/Stripe
0 9 * * 5 | Friday 9 AM — Acharya payout calculation + notification dispatch
0 10 1 1,4,7,10 * | Quarterly: Foundation transfer reminder to Super Admin
2FA enforcement | Mandatory TOTP for ALL admin accounts. Setup forced on first login. No bypass.
IP allowlist | Admin panel accessible only from approved IP addresses — managed in settings
Session timeout | Admin session: 8 hours inactivity -> auto-logout + re-auth required
Failed logins | 5 consecutive failures -> account locked 30 minutes. Alert to Super Admin.
API rate limiting | Admin APIs: 300 req/min per user. Public APIs: 60 req/min. Violations logged.
CSP headers | Strict Content-Security-Policy — prevents XSS in admin panel
Audit retention | All audit logs retained 2 years. Cannot be deleted via UI — DB admin only.
User actions | Edit user profile | Change user membership | Suspend user | Delete user
Financial | Issue refund | Process payout | Record Foundation transfer | Create coupon
AI config | Edit any prompt | Change model | Update token limits | Test prompt
Content | Publish blog post | Approve forecast | Modify rashifal content | Update puja
System | Change API keys | Modify feature flags | Enable maintenance mode | Change roles
Access | Admin login | Impersonation start/end | Failed login attempts | 2FA events
Role | Who Has It | Key Capabilities | Cannot Do
Super Admin | Founder + CTO only | Everything — all 25 modules, AI prompts, payouts, data deletion, role management | Nothing restricted
Admin Manager | Operations head | All modules except: AI prompt editing, bulk data deletion, payout processing | Edit prompts, bulk delete
Content Manager | Blog + SEO team | Blog, Rashifal, Panchang, SEO tools, AI product desc, coupon creation | User PII, payments, AI config
Acharya Coordinator | Acharya onboarding team | Acharya management, consultation monitoring, puja scheduling, GPS certs | Financial data, user PII, AI
Support Agent | Customer support | User view read-only, ticket management, order status view, refund < Rs 500 | Edit users, financial reports
Finance Manager | Accountant / CFO | Orders, payments, GST reports, Acharya payouts, Foundation transfers | User PII, content, AI config
Acharya (Self-panel) | Each Acharya individually | Own review queue, own profile, own schedule, own earnings dashboard | Any other user's data
Setup Commands (one-time)
# Install Filament Shield
php artisan shield:install

# Generate policies for all Filament resources
php artisan shield:generate --all

# Make first user Super Admin
php artisan shield:super-admin --user=1

# Create roles and assign permissions via Admin Panel UI
# No additional code needed — managed from Settings > Roles
Table | Primary Key | Key Columns | Relationships
users | UUID | name,email,phone,dob,tob,pob_lat/lng/tz,rashi,nakshatra,lagna,whatsapp_number,badges(JSON),membership_id | -> subscriptions,orders,kundalis
kundalis | UUID | user_id,chart_data(JSON),lagna,rashi,nakshatra,dasha_data(JSON) | -> users
orders | UUID | user_id,order_type,amount,currency,razorpay_order_id,stripe_payment_id,status | -> users,forecasts
forecasts | UUID | order_id,user_id,plan_type,language,chart_data(JSON),content(JSON chapters),pdf_s3_key,status,verified_by,verification_note | -> orders,acharyas
acharyas | UUID | name_hi,name_en,slug,photo_url,video_intro_url,specializations(JSON),rating_avg,is_verified,payout_upi,consultation_enabled | -> bookings,puja_bookings
bookings | UUID | user_id,acharya_id,duration_min,scheduled_at,agora_channel,status,user_topic,session_notes,recording_url | -> users,acharyas
puja_bookings | UUID | user_id,puja_slug,tier,acharya_id,scheduled_at,sankalp_data(JSON),model,status,gps_coordinates(JSON),certificate_url,prasad_tracking(JSON) | -> users,acharyas
subscriptions | UUID | user_id,plan,billing_cycle,amount,currency,status,razorpay_sub_id,stripe_sub_id,whatsapp_number,whatsapp_opted_in,notification_prefs(JSON),pause_until | -> users
products | UUID | name_hi,name_en,slug,category,price,planet,mukhi,caution_level,consult_required,energized,qr_code,stock,energization_video_url | —
shop_orders | UUID | user_id,items(JSON),amount,status,shipping_address(JSON),courier,tracking_number | -> users
ai_prompts | UUID | name,model,system_prompt(encrypted),user_prompt_template,max_tokens,temperature,version,is_active,created_by | —
ai_logs | bigint | prompt_name,model,input_tokens,output_tokens,cost_usd,response_time_ms,status,error_message,created_at | —
foundation_transfers | UUID | quarter,year,total_revenue,net_profit,transfer_amount,bank_proof_s3_url,acknowledgment_s3_url,transferred_at | —
admin_users | bigint | name,email,password(bcrypt),role,two_factor_secret,last_login_at,last_login_ip | —
audit_logs | bigint | admin_user_id,action,resource_type,resource_id,old_values(JSON),new_values(JSON),ip_address,user_agent,created_at | -> admin_users
webhook_logs | bigint | gateway,event_type,payload(JSON encrypted),signature_valid,processed_at,processing_status,error | —
Endpoint | Method | Auth Role | Purpose
GET /admin/api/dashboard | GET | Any Admin | Dashboard KPI data — all counters
GET /admin/api/ai/logs | GET | Super Admin | AI call logs with cost breakdown
PUT /admin/api/ai/prompts/{id} | PUT | Super Admin | Update AI prompt — versioned
POST /admin/api/ai/prompts/test | POST | Super Admin | Live test prompt with sample chart JSON
POST /admin/api/forecasts/{id}/regenerate | POST | Admin Manager | Re-trigger Claude generation for failed forecast
POST /admin/api/forecasts/{id}/approve | POST | Acharya + Admin | Approve forecast — triggers PDF generation
POST /admin/api/pujas/{id}/gps | POST | Admin + Coordinator | Record GPS coordinates for puja
POST /admin/api/foundation/transfer | POST | Finance + Super Admin | Record quarterly Foundation transfer
POST /admin/api/isr/revalidate | POST | Admin Manager | Trigger Next.js ISR page revalidation
GET /admin/api/analytics/revenue | GET | Finance + Admin | Revenue report — date range + breakdown
POST /admin/api/whatsapp/broadcast | POST | Admin Manager | Send WhatsApp broadcast to members
GET /admin/api/queue/stats | GET | Super Admin | Horizon queue statistics
POST /admin/api/users/{id}/impersonate | POST | Super + Support | Start user impersonation session
DELETE /admin/api/users/{id}/impersonate | DELETE | Super + Support | End impersonation session
POST /admin/api/payments/{id}/refund | POST | Finance + Support | Issue full or partial refund
Job Class | File Path | Queue | Max Attempts
ForecastGenerationJob | Jobs/ForecastGenerationJob.php | high | 1 (no retry — manual)
ClaudeChapterJob | Jobs/ClaudeChapterJob.php | high | 3 (auto-retry)
GeminiRashifalJob | Jobs/GeminiRashifalJob.php | high | 2
PanchangGenerationJob | Jobs/PanchangGenerationJob.php | high | 2
PDFGenerationJob | Jobs/PDFGenerationJob.php | default | 2
EmailDeliveryJob | Jobs/EmailDeliveryJob.php | low | 3
WhatsAppBroadcastJob | Jobs/WhatsAppBroadcastJob.php | default | 2
WebhookProcessJob | Jobs/WebhookProcessJob.php | high | 3
ISRRevalidationJob | Jobs/ISRRevalidationJob.php | default | 2
FoundationReportJob | Jobs/FoundationReportJob.php | low | 1
AcharyaPayoutCalcJob | Jobs/AcharyaPayoutCalcJob.php | low | 1
GPSCertificateJob | Jobs/GPSCertificateJob.php | default | 2
PrasadDispatchJob | Jobs/PrasadDispatchJob.php | low | 1
UserBadgeCheckJob | Jobs/UserBadgeCheckJob.php | low | 1
Volume Scenario | Claude Cost | Gemini + OpenAI | Total Monthly
Launch: 100 forecasts/day | 100 x Rs 78 = Rs 7,800 | Rashifal + Blog + Support = Rs 940 | ~Rs 8,740/month
Growth: 500 forecasts/day | 500 x Rs 78 = Rs 39,000 | Same = Rs 940 | ~Rs 39,940/month
Scale: 2000 forecasts/day | 2000 x Rs 78 = Rs 1,56,000 | Same = Rs 940 | ~Rs 1,56,940/month
Daily alert | Rs 5,000/day — email to Super Admin + Slack notification
Weekly alert | Rs 25,000/week — email to Super Admin + Finance Manager
Monthly budget | Rs 1,00,000/month — if approaching, some tasks auto-downgrade to cheaper model
Per-report ceiling | Rs 200 maximum per individual forecast — if Claude job exceeds: fail gracefully + alert
Anomaly detection | If any single hour costs > Rs 2,000 — immediate alert — possible abuse or misconfiguration