Introduction

What This Application Does

Ignitech AI Caller is an automated outbound calling system for lead qualification. It receives leads from sources such as IndiaMART, Facebook Lead Ads, Facebook Page messages, and CSV uploads; stores them in PostgreSQL; calls valid phone numbers through LiveKit SIP using a Vobiz trunk; runs an AI voice agent; saves the structured call outcome; and pushes qualified leads into Talking Shops CRM. The application is built for sales teams that need a repeatable first qualification call before a human follows up.

Main Outcomes

New leads are collected into one leads table.
The scheduler calls leads only during configured calling days and hours.
A voice agent speaks with the lead and collects qualification answers.
Each call attempt is stored in the calls table with status, duration, failure reason, and qualification JSON.
Qualified calls are pushed to Talking Shops as follow-up tasks.
Rejected, unanswered, stale, or failed calls can be retried with limits.
Streamlit dashboards show lead pipeline, call status, pushed records, and call-minute usage.

Core Technologies

Tool	Use in this application
Python	Main runtime for all services
LiveKit Agents	Voice-agent runtime and room/session handling
LiveKit SIP	Connects phone calls to LiveKit rooms
Vobiz	SIP provider used to dial and receive PSTN phone calls
Sarvam AI	STT, LLM, and TTS stack for the Sarvam agent mode
OpenAI Realtime	Low-latency realtime voice conversation mode
PostgreSQL	Persistent lead, call, retry, and qualification storage
asyncpg	Async PostgreSQL driver
aiohttp/httpx	HTTP servers and API integrations
Streamlit	Internal lead pipeline and call-minute dashboards
Docker Compose	Production-style multi-service runtime
Talking Shops	CRM destination for qualified follow-up tasks
IndiaMART API	Polling lead source
Facebook Graph/Webhooks	Lead Ads and Page message lead source

Service Map

Lead sources
  -> lead_fetcher.py / Streamlit CSV upload
  -> PostgreSQL leads table
  -> scheduler.py
  -> LiveKit SIP outbound participant
  -> Vobiz trunk / phone network
  -> LiveKit room: ignitech-lead-{lead_id}
  -> agent.py or agent_realtime.py
  -> PostgreSQL calls table
  -> talkingshops_pusher.py
  -> Talking Shops CRM

Repository Layout

File	Purpose
`agent.py`	Sarvam-based LiveKit voice agent
`agent_realtime.py`	OpenAI Realtime-based LiveKit voice agent
`db.py`	PostgreSQL schema, connection pool, and DB helper functions
`lead_fetcher.py`	IndiaMART poller and Facebook webhook server
`scheduler.py`	Lead polling, calling-window enforcement, SIP call trigger, CRM push loop
`sip_setup.py`	LiveKit SIP trunk setup, listing, deletion, and manual test calls
`talkingshops_pusher.py`	Formats and pushes qualified calls to Talking Shops
`streamlit_app.py`	Internal pipeline dashboard and CSV lead upload
`pages/minutes.py`	Streamlit page for successful call minutes
`token_server.py`	LiveKit token server for browser/local testing
`docker-compose.yml`	Database, agent, fetcher, scheduler, dashboard, and token server services
`.env.example`	Full configuration template

Getting Started

Architecture

Integrations

Operations

What This Application Does

Main Outcomes

Core Technologies

Service Map

Repository Layout

​What This Application Does

​Main Outcomes

​Core Technologies

​Service Map

​Repository Layout

What This Application Does

Main Outcomes

Core Technologies

Service Map

Repository Layout