# ⚡ Synapse: Advanced Multi-Query AI Assistant

Synapse is a high-performance **Retrieval-Augmented Generation (RAG)** application built to extract precise, well-organized information from PDF documents. By utilizing **Multi-Query Expansion**, Synapse overcomes the limitations of standard keyword search, ensuring it finds the most relevant information even if the user's question is vague.



## 🌟 Key Features

* **Multi-Query Expansion:** Uses Groq's Llama 3.3 to rewrite user queries into multiple variations, increasing retrieval accuracy (Recall).
* **Persistent Cloud Memory:** Integrated with **Neon DB (PGVector)** to store document embeddings permanently in the cloud.
* **Explainable AI:** Provides direct **Source Citations** including page numbers and exact text snippets for every answer.
* **Context-Aware Chat:** Maintains a session-based history to handle follow-up questions (e.g., "What about the second one?").
* **Sub-Second Latency:** Powered by **Groq LPU** inference for near-instant responses.

## 🛠️ Tech Stack

* **Frontend:** [Streamlit](https://streamlit.io/)
* **Orchestration:** [LangChain](https://www.langchain.com/)
* **Brain (LLM):** Groq (Llama 3.3-70B)
* **Vector Database:** Neon DB (PostgreSQL + PGVector)
* **Embeddings:** Hugging Face (`all-MiniLM-L6-v2`)

## 📂 Project Structure

```text
Synapse-Cloud/
├── .streamlit/             # Streamlit configuration
│   └── secrets.toml        # Local secrets (ignored by Git)
├── app.py                  # Main application logic
├── requirements.txt        # Python dependencies
├── packages.txt            # Linux system dependencies (libpq-dev)
└── .gitignore              # Files to exclude from GitHub