🧠 Brainwonders Parent Assistant

Empowering Parents with Instant Answers on Brainwonders Programs

This repository hosts the code for an intelligent AI-powered assistant designed to help parents quickly find information about Brainwonders' career counselling programs, pricing, and related services. Leveraging advanced Retrieval-Augmented Generation (RAG) techniques, this assistant provides accurate, concise, and context-aware responses based on a curated knowledge base of Brainwonders' offerings.

---

🚀 Live Demo

Experience the Brainwonders Parent Assistant live on Render:

Brainwonders Parent Assistant

✨ Features

• Intelligent Q&A: Answers parent queries about Brainwonders' programs, pricing, methodologies (like DMIT), and services.
• Context-Aware Responses: Utilizes RAG to fetch relevant information from a dedicated knowledge base, ensuring answers are grounded in provided data.
• Multi-Document Support: Capable of ingesting and retrieving information from various document formats, including:
  • Markdown files (.md)
  • Plain text files (.txt)
  • PDF documents (.pdf)
• Chat History Integration: Maintains conversational context, allowing for follow-up questions and refined answers based on the ongoing dialogue.
• Streaming Responses: Provides a smoother user experience by streaming AI responses word-by-word.
• User-Friendly Interface: Built with Streamlit for an interactive and accessible chat interface.
• Scalable Knowledge Base: Uses ChromaDB as a local vector store for efficient semantic search, pre-built and committed for faster deployment.
• Robust Error Handling: Includes basic error handling for API calls and document loading.

---

💡 How It Works (Technical Overview)

The Brainwonders Parent Assistant is built upon a Retrieval-Augmented Generation (RAG) architecture using the LangChain framework, powered by Google's Gemma LLM and Embeddings.

1. Data Ingestion & Indexing:

   • The data/ directory serves as the knowledge base, containing Markdown, text, and PDF documents.
   • DirectoryLoader from langchain-community is used to load all supported documents.
   • RecursiveCharacterTextSplitter breaks down these documents into smaller, manageable chunks.
   • GoogleGenerativeAIEmbeddings convert these text chunks into numerical vector embeddings.
   • ChromaDB is used as the vector store to efficiently index and store these embeddings locally. This chroma_db directory is pre-built during development and committed to the repository for quick deployment startup.

2. Retrieval:

   • When a user submits a query, Chroma's retriever (powered by GoogleGenerativeAIEmbeddings) performs a semantic similarity search against the indexed vector store.
   • It retrieves the top k (set to 3) most relevant document chunks that semantically match the user's question.

3. Generation:

   • A ChatGoogleGenerativeAI model (specifically gemma-3-27b-it) acts as the Large Language Model (LLM).
   • A ChatPromptTemplate is crafted to combine:
     • The user's current question.
     • The retrieved relevant context from the documents.
     • The ongoing chat history.
     • System instructions to guide the LLM's behavior (e.g., sticking to the context, handling general queries, prompting for follow-up).
   • create_stuff_documents_chain then feeds this combined information to the LLM, which generates a comprehensive and relevant answer.
   • StreamlitChatMessageHistory ensures the conversation history persists across interactions, allowing the LLM to maintain context.

4. User Interface:

   • Streamlit provides the interactive web interface, handling chat input, displaying messages (including streaming AI responses), and managing session state for a seamless user experience.

---

🛠️ Technologies Used

• Python 3.9+
• Streamlit: For building the interactive web application.
• LangChain: Framework for building LLM applications, managing RAG pipeline, and conversational history.
  • langchain-core
  • langchain-community (for document loaders and chat history)
  • langchain-chroma (for vector store)
  • langchain-google-genai (for Google LLM and Embeddings)
• Google Gemma (gemma-3-27b-it): The Large Language Model used for text generation.
• Google Generative AI Embeddings (embedding-001): For converting text into vector representations.
• ChromaDB: Lightweight, in-memory (or persisted locally) vector database for semantic search.
• python-dotenv: For managing environment variables securely.
• pypdf: For processing PDF documents.

---

⚙️ Setup and Installation (Local)

To run this project locally, follow these steps:

1. Clone the repository:

   git clone [https://github.com/JoshPola96/Brainwonders-Parent-RAG-QA.git](https://github.com/JoshPola96/Brainwonders-Parent-RAG-QA.git)
   cd Brainwonders-Parent-RAG-QA

2. Create a Python virtual environment (recommended):

   python -m venv venv
   # On Windows:
   .\venv\Scripts\activate
   # On macOS/Linux:
   source venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

4. Set up Google API Key:

   • Create a .env file in the root of your project directory.
   • Add your Google Generative AI API key to it:

     GOOGLE_API_KEY="your_google_api_key_here"
     

   • You can obtain a Google API Key from the Google AI Studio: https://aistudio.google.com/app/apikey

5. Prepare the Data:

   • Ensure your data/ directory contains your brainwonders_info.md and any other .txt or .pdf documents you wish to include in the knowledge base.
   • Run the app once locally to generate the chroma_db directory:

     streamlit run app.py

     (This will create the chroma_db directory containing your vector store. This directory should be committed to Git for cloud deployment.)

6. Run the Streamlit app:

   streamlit run app.py

   The app will open in your default web browser.

---

☁️ Deployment on Render

This application is designed for easy deployment on platforms like Render.

1. Prepare your GitHub Repository:

   • Ensure all necessary files (app.py, requirements.txt, data/ directory with documents, and the pre-built chroma_db/ directory) are committed to your GitHub repository.
   • Crucially, your chroma_db/ directory must be committed to Git. This ensures your vector store is persisted and available upon deployment.

2. Render Setup:

   • Go to render.com and create a new Web Service.
   • Connect your GitHub repository.
   • Configure the service details:
     • Build Command: pip install -r requirements.txt
     • Start Command: streamlit run app.py --server.port $PORT --server.enableCORS false --server.enableXsrfProtection false
     • Environment Variables: Add GOOGLE_API_KEY with your actual API key.

3. Deploy: Render will automatically build and deploy your application.

---

📧 Contact

For any questions or feedback, please contact [Joshua Peter/josh19peter96@gmail.com].