🚀 Your Personal AI Knowledge Assistant powered by RAG + Hybrid Intelligence
An advanced AI system that doesn’t just store information — it understands, retrieves, and intelligently responds using both your personal knowledge and general intelligence.
-
📝 Smart Note Storage
Store and manage your personal knowledge -
🤖 AI Q&A (Hybrid RAG System)
- Answers from your notes (RAG-based)
- Falls back to general knowledge if needed
- Clearly mentions when answer is not from notes
-
🧠 Semantic Search with Embeddings
Uses Sentence Transformers for meaning-based retrieval -
📄 Document Understanding (PDF / DOCX / TXT)
Upload files → auto chunking → ask questions -
✨ AI Summarization
Get concise summaries of notes and documents -
🎯 Multi-Question Handling
Handles multiple questions in one query -
📊 System Stats
Tracks notes count and AI usage
Frontend
- Next.js
- TypeScript
- Tailwind CSS
Backend
- FastAPI
Database
- MongoDB
AI / ML
- Ollama (LLaMA3)
- Sentence Transformers (Embeddings)
- Cosine Similarity (Vector Search)
- Notes & documents stored in MongoDB
- Each note converted into vector embeddings
- User query → embedding
- Compared with stored notes using cosine similarity
- Top relevant notes selected dynamically
- Context created from best matches
- If answer exists in notes → uses notes
- If not → uses general knowledge
- Adds transparency: "This information is not from your personal notes."
- Returns answer
- Shows sources only when relevant
I designed and developed this project end-to-end, including:
- The responsive frontend using Next.js, TypeScript, and Tailwind CSS
- Backend APIs using FastAPI and Python
- MongoDB integration for storing notes and document information
- PDF, DOCX, and TXT document-upload workflows
- Text extraction, chunking, and embedding generation
- Semantic search using Sentence Transformers and cosine similarity
- A RAG pipeline for retrieving relevant personal context
- Hybrid response logic for personal and general knowledge
- AI-powered note and document summarization
- Transparent fallback messaging when an answer is not supported by the user's notes
- Light and dark user-interface experiences
The most challenging part was determining when the system had enough relevant context to answer from the user's documents and when it should fall back to general knowledge.
I addressed this by implementing similarity-based retrieval and clear fallback messaging so users can understand whether an answer came from their personal knowledge base.
- Combines personal knowledge retrieval with general AI intelligence
- Uses semantic search instead of basic keyword matching
- Hybrid system (RAG + General Knowledge)
- Semantic search (not just keyword matching)
- Handles edge cases intelligently
- Designed like a real-world AI product
git clone https://github.com/mittalsonal/AI-Second-Brain.git
cd ai-second-braincd frontend
npm install
npm run devcd backend
pip install -r requirements.txt
uvicorn app.main:app --reload- RAG-based context-aware answering
- Semantic search with vector embeddings
- Notes and document understanding
- Hybrid personal and general knowledge responses
- Context-aware responses
- Notes + PDF understanding system
- End-to-end full-stack AI implementation
- Add authentication and separate knowledge bases for individual users
- Add support for more document formats and larger files
- Introduce a dedicated vector database
- Add retrieval-quality evaluations
- Improve source citations at the document-chunk level
- Add conversation memory and knowledge organization
- Deploy the complete AI pipeline for public use
Sonal Mittal
Full Stack Developer and AI Engineer
- GitHub: mittalsonal
- Email: sonalmittal140103@gmail.com
- LinkedIn : https://www.linkedin.com/in/mittalsonal/
- Portfolio : https://sonalmittal.vercel.app/
⭐ If you find this project useful, consider giving it a star!