LeadGen AI

August 12, 2026 · View on GitHub

FastAPI SaaS for small Indian local businesses. Live: https://leadsgenai.in (Hostinger VPS, Mumbai).

Two separate products, sold as distinct SKUs:

Product	What it is	Price
AI Automated Marketing	Done-for-you social-media + local-lead marketing (programmatic content, posting, lead capture, CRM follow-up)	Main ₹1,999/mo · Combo/Advanced ₹5,999/mo
AI Voice Calling Agent	Standalone full AI telecaller (inbound auto-callback + DLT-gated outbound)	Flat per niche-band: A ₹4,999/mo · B ₹9,999/mo · C ₹19,999/mo

Money path: free lead magnets (/audit, /site-audit, /demo) + pSEO + auto email-outreach → inquiry → /pricing → /start → manual UPI (owner-confirmed) → subscription + top-up minute packs. Manual UPI is the canonical and only payment rail (owner-verified).

The entire AI stack runs on free providers only (no paid STT/TTS/LLM).

Products

1. AI Automated Marketing

AI social media content (posts/reels copy, Hinglish-optimised) per the client's niche
Programmatic SEO + mini-sites (/b/{slug})
Lead capture widget + Hot Queue (/app/inbox) for follow-up
AI email/WhatsApp outreach (ban-safe: cold WhatsApp OFF, 1-click human send)
Plans:
- main ₹1,999/mo
- advanced ₹5,999/mo (includes AI-callback feature, 500 min)
- growth ₹2,999 (legacy hidden)
- ₹0 7-day trial (marketing-lite)
- Annual = 10x monthly (2 months free)
- Top-ups: 100/250/500 min = ₹1,499/₹3,499/₹5,999

2. AI Voice Calling Agent

Full AI telecaller: STT → LLM → TTS over real phone calls (Vobiz India SIP)
Compliance-gated: DND scrub (fail-closed), TRAI window 9am–7pm, AI disclosure at call start, consent ledger, DLT for cold outbound, 90-day recording retention
Standalone pricing per niche band: voice_a_monthly ₹4,999 · voice_b_monthly ₹9,999 · voice_c_monthly ₹19,999

Architecture

Internet ──> Caddy (host, TLS leadsgenai.in) ──> leadgen_app :8000 (FastAPI)
                                                    ├── Postgres leadgen_db (via PgBouncer :6432)
                                                    ├── Redis leadgen_redis :6379 (Broker + state)
                                                    ├── Qdrant :6333 (RAG: single kb_main, namespaced)
                                                    ├── leadgen_worker (Celery) + leadgen_scheduler
                                                    ├── FreeSWITCH + WS voice stream (L16/16k)
                                                    └── Obs stack (Prometheus/Grafana/Loki/...)

Backend: FastAPI (async), domain routers in app/api/ (split by domain), engines in app/platform/, voice in app/voice_agent/ + app/telephony/, billing in app/billing/
Frontend: server-rendered HTML in frontend/ (marketing site, Mission Control /app/automation, HQ /app/office, 4 dashboards)
Agents: coordinator / dag_engine / process_engine / self_improve loops in app/agents/, governed harness in app/agents/harness/ (INERT by default)

AI Stack (all free)

Component	Provider
LLM (primary)	Mistral `mistral-small-latest`
LLM (fallback)	Groq · Cerebras · NVIDIA NIM · SambaNova · OpenRouter
Voice-scoped LLM	Gemini (9-key rotation pool)
STT	Groq `whisper-large-v3` (primary) · Gemini audio (fallback)
TTS	EdgeTTS `hi-IN-SwaraNeural` (free)
Images/video	Pollinations
Embeddings	fastembed `multilingual-e5-small`
Telephony	Vobiz (India SIP) · Twilio (international fallback)
Email	Hostinger SMTP/IMAP (`admin@leadsgenai.in`)
Prospecting	Google Maps Places (New)
Web search	SearXNG (self-host)
Push	ntfy (self-host)
WhatsApp	Meta Cloud + own WAHA :3111
Payments	Manual UPI only (`UPI_VPA`)

Deploy

Single canonical deploy entrypoint: scripts/deploy_vps.sh (mandatory APP_VERSION=<sha>, /health.version gate, 5-service skew check). Full runbook: memory/playbooks.md.

Manual: git push origin main → SSH to VPS → cd /opt/leadgen && setsid nohup bash scripts/deploy_vps.sh > /tmp/dep.log 2>&1 & → poll /tmp/dep.log.

docker-compose.vps.yml is the only production compose file. Dockerfile.lock is the only build input.

Quick Start (dev, Windows, py3.12)

python -m venv .venv
.venv\Scripts\pip install --no-deps -r requirements.lock.txt
.venv\Scripts\python.exe -m uvicorn app.main:app --reload --port 8000

Note: Run dev is [UNVERIFIED locally] — import verified via prod_check.

Tests (targeted): .venv\Scripts\python.exe -m pytest tests/test_billing_truth_2026.py -q
Full suite: scripts\run_tests.bat (then read pytest_run.log)
Lint: .venv\Scripts\python.exe -m ruff check app
Verify gate: .venv\Scripts\python.exe scripts\prod_check.py and scripts\check_secrets.py

Repo map

app/            FastAPI backend (routers, engines, voice, billing, agents)
frontend/       Server-rendered HTML (marketing site + app dashboards)
alembic/        DB migrations (DB_CREATE_ALL=0 → Alembic-only in prod)
tests/          pytest suite
scripts/        Deploy + ops (deploy_vps.sh = canonical deploy)
docs/           Architecture, ADRs (docs/adr/), runbooks, context
memory/         Decisions/backlog/incidents/playbooks (append-only)
monitoring/     Prometheus rules, gatus synthetic checks
deploy/         Non-production compose + scheduler config
docker-compose.vps.yml   CANONICAL production stack
Dockerfile.lock          CANONICAL build input

Compliance

TRAI/telecom: DND scrub fail-closed, AI disclosure at call start, promo window 9am–7pm, consent-ledger opt-out, no foreign trunks for India-domestic, cold auto-calls require DLT
DPDP Act 2023: purpose limitation, data minimisation, consent basis, 90-day recording retention, purge API + Grievance Officer in /privacy, strict customer isolation
Secrets: .env only (gitignored), never committed

Docs

Operating manual: CLAUDE.md / AGENTS.md (byte-identical)
Loop ledger: progress.md
Architecture + ADRs: docs/
Knowledge base: memory/ (memory/INDEX.md first)