AutoStream AI Agent: An Intelligent Lead Capture Workflow Based on LangGraph

AutoStream AI Agent is an intelligent conversational agent system for video editing SaaS products. It converts social media chats into high-quality sales leads through intent recognition, RAG retrieval, and automated lead capture. With LangGraph as its core framework, the system builds a multi-node workflow to achieve seamless conversion from social conversations to sales leads, creating tangible business value for enterprises.

In the digital marketing environment, enterprises face the challenge of converting social media interactions into sales leads. Traditional customer service systems respond passively and lack the ability to proactively identify high-intent customers. AutoStream AI Agent is designed specifically to address this pain point. Combining large language model intent recognition, RAG technology, and a structured lead capture workflow, its core value lies in being an intelligent agent with clear business goals—understanding user intent, providing accurate information, guiding users to leave contact details, and creating business value.

The system uses LangGraph (an extension of the LangChain ecosystem) to build a multi-node agent workflow, which includes three core nodes: 1. Intent Classifier: Classifies user messages into greeting, product inquiry, and high-intent categories; 2. RAG Retriever: For product inquiry messages, retrieves relevant fragments from the knowledge base and generates accurate answers; 3. Lead Manager: When a high-intent user is detected, guides them to provide information such as name and email, then verifies and stores it. The system manages conversation states (history, intent, leads, node position) through the AgentState object to maintain context coherence.

Functional Features: 1. Intelligent Intent Recognition: A three-level classification system that adopts different strategies for different intents; 2. RAG-driven Q&A: Understands natural language questions, combines multiple document fragments to generate comprehensive answers, and maintains timeliness; 3. Multi-turn Conversation Memory: Remembers user needs and collects leads in multiple turns to reduce user burden; 4. Controlled Tool Execution: The mock_lead_capture tool supports switching between test and production environments to control behavior boundaries.

Application Example: A user inquires about pricing → The system identifies it as a product inquiry and calls RAG to display package comparisons → The user expresses interest in the Pro plan → The system identifies high intent and activates the Lead Manager → Collects name, email, and usage platform → Confirms successful lead capture.

Technology Stack: Python 3.9+ (a mainstream AI language that supports modern features); LangChain + LangGraph (interfaces for interacting with large models, state management, and workflow orchestration); Streamlit (quickly builds UI prototypes); JSON knowledge base (easy to manage in the prototype phase, can be migrated to a vector database in the future).

Project Structure: src/ (core business logic), app/ (Streamlit UI), data/ (JSON knowledge base), main.py (CLI entry), which embodies the principle of separation of concerns.

Future Expansion: Database integration (persistent storage of leads and conversations); Email verification (improves lead quality); UI button interaction (reduces user input costs); WhatsApp integration (expands reach channels); Cloud deployment (high availability and elastic scaling).

Insights for Developers: Start simple (quickly validate ideas); Workflow thinking (treat as a state machine); Intent-driven design (personalized responses); Human-machine collaboration (AI handles conversations, human systems process key operations); Progressive enhancement (path from prototype to production).

AutoStream AI Agent is a well-designed intelligent conversational system prototype that demonstrates how to encapsulate large language model capabilities into an agent application with clear business goals. Through the organic combination of intent recognition, RAG retrieval, and lead capture, it provides enterprises with an automated solution for converting social interactions into sales leads. For AI agent developers, it is a valuable open-source project with clear code structure, reasonable technology selection, and a clear expansion path, which can serve as an entry-level case for LangGraph or a starting point for production systems.