Synapse 🧠

An intelligent, self-correcting Retrieval-Augmented Generation (RAG) pipeline built with Python, LangChain, and Google's Gemini API.

This project goes beyond a standard RAG implementation. It features an active learning loop that enables the system to learn from its mistakes and improve its knowledge base over time.

---

Key Features

• Retrieval-Augmented Generation (RAG): Answers user questions by retrieving relevant context from a local document set.
• Self-Correcting Knowledge Base: When a user asks a question and no relevant documents are found, the system automatically:
  1. Logs the "missed" query to a text file.
  2. Embeds the user's query text.
  3. Adds the query as a new document back into the vector store.
• Constantly Evolving: Because of this "active learning" feature, the application's knowledge base grows with every user interaction, allowing it to answer similar queries correctly in the future.

---

Technology Stack

• Language: Python
• Core Framework: LangChain
• LLM: Google Gemini (via langchain-google-genai)
• Vector Store: ChromaDB
• Embeddings: Sentence Transformers (all-MiniLM-L6-v2)

---

How It Works

1. Ingestion: All documents in the /files directory are loaded, chunked, and embedded into a ChromaDB vector store.
2. Query: A user asks a question.
3. Retrieve: The system searches the vector store for relevant chunks of text.
4. Evaluate:
   • If Context is Found: The context and the query are sent to the Gemini LLM to generate a factual answer.
   • If No Context is Found: The system logs the query to files/missed_queries.txt, then embeds and adds the query itself to the vector store. This allows it to be found in future, similar searches. The user receives a standard conversational response.
5. Repeat: The application gets smarter with every missed query.

---

🚀 Setup and Installation

1. Clone the Repository

git clone https://github.com/Abhi0049k/Synapse.git
cd Synapse

2. Create a Virtual Environment

uv venv can create the virtual environment and install all packages from your requirements.txt and uv.lock files with just two commands.

# On macOS/Linux
python3 -m venv .venv
source .venv/bin/activate

# On Windows
python -m venv .venv
.\.venv\Scripts\activate

3. Install Dependencies

uv pip sync requirements.txt

4 Set Up Your Environment

This is the most important step. Your API keys must be kept secret.

1. Create a file named .env in the root of the project.
2. Add your Google API key to it:

   GOOGLE_API_KEY=your_secret_api_key_here
   

3. This .env file is already listed in .gitignore, so it will never be uploaded to GitHub.

5. Add Your Documents

Place any PDF, TXT, or other documents you want the RAG to learn from into the /files directory.

---

💡 Usage

Run the main.py script to start the ingestion process and begin the chat session.

uv run main.py

The application will first process all your documents and build the vector store. Once complete, you will be prompted to ask questions:

Pipeline setup successfully!
Please enter what you want to search for or enter exit to end the simulation:

---

📜 License

This project is licensed under the GNU General Public License v3.0. See the LICENSE file for more details.