AutoStream AI Agent – Social-to-Lead Workflow

This project is a Conversational AI Agent built for a fictional SaaS platform AutoStream, designed to convert user conversations into qualified business leads.

It demonstrates a complete agentic workflow using intent detection, RAG (Retrieval-Augmented Generation), state management, and tool execution.

Features:

• Intent Detection (Greeting, Pricing, High-Intent)
• RAG-based Knowledge Retrieval (local JSON)
• Stateful Conversation (multi-turn memory)
• Tool Execution (Lead Capture)
• Gemini LLM Integration for natural responses

Architecture Overview :

This system is built using a modular design:

• Intent Detection

  • Classifies user input into greeting, pricing, high-intent, or general
  • Uses Gemini LLM for better understanding.

• RAG Pipeline

  • Retrieves pricing and policy data from a local JSON knowledge base
  • Ensures accurate and non-hallucinated responses

• State Management

  • Stores user information (name, email, platform)
  • Maintains conversation across multiple turns

• Agent Logic

  • Routes user input based on intent
  • Controls conversation flow and decision-making

• Tool Execution

  • Captures leads only after all required inputs are collected
  • Simulates backend API call

How to run

Make sure Python 3.10+ is installed.

1. Clone my repo:

git clone https://github.com/Ritesh2332/autostreamAgent.git
cd autostreamAgent

2. Create a virtual environment:

python -m venv venv

3. Activate the virtual environment:

venv\Scripts\activate

4. Install the required packages:

pip install -r requirements.txt

5. Run the agent:

python app/main.py

6. Example usage

You: Hi
You: What is your pricing?
You: I want to buy this
You: Your Name
You: your@email.com
You: YouTube