Synapse

Synapse is a minimal agentic GenAI chat application built to demonstrate practical understanding of LangChain, LLM agents, and RAG (Retrieval-Augmented Generation), served via FastAPI with a lightweight vanilla JS frontend.

The project intentionally prioritizes clarity, modularity, and explainability over production-scale complexity.

✨ Key Features

• 🤖 Agentic LLM Chat

  • Built using LangChain’s create_agent
  • Powered by Groq-hosted foundation models. Eg. llama-3.1-8b-instant, openai/gpt-oss-20b
  • Supports short-term conversational memory within a chat

• 📄 RAG (Retrieval-Augmented Generation)

  • Upload PDF or TXT documents
  • Agent can retrieve relevant context from uploaded documents
  • Works seamlessly alongside normal LLM chat

• 🧠 Conversational Memory

  • Agent remembers recent messages in the same chat
  • Memory is scoped to a single session (no global persistence)

• 🧩 Modular Architecture

  • Clean separation between API layer, agent logic, tools, and infrastructure
  • Designed to be easily extended (auth, multi-user, persistence, LangGraph)

• 🌐 Minimal Frontend

  • Vanilla HTML, CSS, and JavaScript
  • Simple chat interface for demonstration and testing

🎯 Purpose of This Project

Synapse exists to demonstrate:

• Agent-based GenAI application design
• Practical LangChain usage
• Clean backend architecture
• Thoughtful engineering tradeoffs

It is intentionally minimal, modular, and extensible.

🏗️ Architecture Overview

graph TD
    %% Node Definitions with Bold Titles
    A["<b>FRONTEND (UI)</b><br/>HTML + CSS + Vanilla JS<br/>• Chat input<br/>• File upload (PDF/TXT)<br/>• Message rendering"]
    
    B["<b>FASTAPI BACKEND</b><br/>• Request validation<br/>• Routing<br/>• Lifecycle management"]
    
    C["<b>LANGCHAIN RUNTIME</b><br/>• Agent executor<br/>• Memory<br/>• Tools (<b>RAG</b>)"]
    
    D["<b>GROQ Foundation Models</b><br/> • LlaMa models<br/> • OpenAI models <br/> + Vector Store (<b>RAG</b>)"]

    %% Flow Connections with bold labels
    A ==>|<b>&nbsp; HTTP JSON / Multipart &nbsp;</b>| B
    B ==> C
    C ==> D

    %% Styling for High Visibility
    %% Using high-contrast borders and semi-transparent fills 
    %% that work in both Light and Dark modes
    style A fill:#3498db22,stroke:#3498db,stroke-width:3px,rx:10,ry:10
    style B fill:#9b59b622,stroke:#9b59b6,stroke-width:3px,rx:10,ry:10
    style C fill:#2ecc7122,stroke:#2ecc71,stroke-width:3px,rx:10,ry:10
    style D fill:#e67e2222,stroke:#e67e22,stroke-width:3px,rx:10,ry:10

    %% Edge styling
    linkStyle default stroke:#888,stroke-width:2px

Loading

🧠 How RAG Works in Synapse

1. User uploads a PDF or TXT file
2. Document is loaded and split into chunks
3. Chunks are embedded and stored in an in-memory vector store
4. RAG is exposed to the agent as a tool
5. During chat:
   • Agent decides whether to call the RAG tool
   • Retrieved context is injected into the reasoning process
   • Final answer is generated by the LLM

If no document is uploaded, the agent behaves like a normal conversational LLM.

💾 Memory Model

• Short-term, in-process memory
• Maintains recent conversation turns
• Scoped to a single chat session
• No persistence across restarts

This design keeps behavior predictable while remaining extensible to:

• Redis / DB-backed memory
• LangGraph checkpoints
• Multi-user isolation (future scope)

🚫 Out of Scope (Intentional)

• Authentication & authorization
• Multi-user support
• Persistent storage
• Streaming responses
• Advanced UI frameworks

These are intentionally excluded to keep the project focused and explainable.

▶️ Running the Application (Local)

1. Clone the repository
2. Create a virtual environment
3. Install dependencies
4. Add Groq API key to .env as GROQ_API_KEY=...
5. Start the FastAPI server
6. Open the frontend in a browser

(Exact commands will be added once implementation is complete.)

---

🔮 Future Extensions

• Multi-user support with authentication
• Persistent memory using Redis or database
• LangGraph-based orchestration
• Streaming responses
• Tool expansion (web search, code execution, etc.)