🧠 Mental Health RAG Chatbot (Gemini + LangChain + Chroma)

A lightweight mental-health conversational AI assistant built using:

• Google Gemini 2.0 Flash
• LangChain
• ChromaDB
• Retrieval-Augmented Generation (RAG)
• Conversation Memory (last 4 turns = 8 messages)
• Safety filters (anti-jailbreak + topic restriction)

This chatbot ONLY talks about emotional well-being and blocks unsafe or unrelated topics.

---

✨ Features

🔹 1. Retrieval-Augmented Generation (RAG)

Fetches relevant answers from your mental-health dataset stored in ChromaDB.

🔹 2. Conversation Memory

Remembers the last 4 conversation turns (8 messages total), enabling more natural and contextual replies.

🔹 3. Safety Guardrails

Prevents harmful or manipulative prompts like:

• ignore previous
• jailbreak
• switch role
• system override
• bypass

Also blocks off-topic questions politely.

🔹 4. Text Summarization

RAG chunks are summarized before generating the final answer for better clarity.

---

📁 Project Structure

│── main.py             → Chatbot logic + safety + memory + RAG
│── rag_pipeline.py     → ChromaDB retriever
│── ingest.py           → CSV → chunks → embeddings → Chroma
│── system_prompt.py    → Base system instructions
│── data.csv            → Your mental-health FAQ dataset
│── README.md

---

🧠 Workflow (How It Works)

1. User enters a question
       ↓
2. Bot checks: Is it related to mental health?
       ↓
3. Retrieves relevant chunks from ChromaDB
       ↓
4. Summarizes retrieved chunks using Gemini
       ↓
5. Final prompt = Summary + Memory + User Query
       ↓
6. Gemini generates a safe, supportive response
       ↓
7. Conversation memory updates (max 4 turns)