Most PDFs are static and hard to interact with. We wanted to build something where instead of reading documents, you can talk to them.

As students trying to learn, we found that most AI PDF tools either cost money or feel painfully slow, so we built something better:

Basically: your own personal AI tutor, no paywall, no lag.

• 📄 Upload & index any PDF instantly
• 💬 Ask questions in natural language
• 🎯 Context-aware answers grounded in your document, not hallucinated
• 🔍 Source transparency, every answer links back to the exact page and chunk it came from
• ⚡ No fine-tuning needed, just upload and go

PDF Upload → Text Extraction → Chunking → Embeddings (Vector Store)
                                                        ↓
User Query → Query Embedding → Semantic Search → Top Chunks
                                                        ↓
                                          Reranking + LLM Generation
                                                        ↓
                                         Answer + Source References

Traditional LLMs answer from their training data, which is static, potentially outdated, and unaware of your private documents. RAG solves this by:

• Grounding responses in your actual content
• Citing sources so you can verify every answer
• Staying current, no retraining needed when documents change

Think of it like a smart librarian: you ask a question, it finds the most relevant pages, and explains the answer using exactly those pages.

knowpage/
├── frontend/          # Next.js app (UI, PDF viewer, chat)
└── backend/
    ├── app/           # FastAPI server (REST API, PDF management)
    └── ml/            # ML service (embeddings, retrieval, generation)

Browser (Next.js)
      │
      ▼
FastAPI Backend (:8000)  ──────►  ML Service (:8001)
      │                           (embed, retrieve, rerank, generate)
      ▼
   SQLite

• Node.js 18+
• Python 3.10+
• pip
• ollama with tinyllama

git clone https://github.com/gyawaliaadim/KnowPage
cd KnowPage

cd frontend
npm install
npm run dev

# In a new terminal, from the project root
cd backend
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

# In a new terminal, from the project root
cd backend
uvicorn ml.main:app --port 8001

ollama run tinyllama

⚠️ All three services must be running simultaneously. Once up, open http://localhost:3000 in your browser.

In ./backend/app/core/config.py, change:

LLM_URL = "http://ollama:11434"
MODEL_URL = "http://ml:8001"

docker compose up
# Once running, stop it to install the model
docker compose stop

docker ps
docker exec -it knowpage-ollama-1 ollama pull tinyllama

docker exec -it knowpage-backend-1 python -m app.scripts.init_db

docker compose up

• Handling large PDF files efficiently
• Designing a clean upload → process pipeline
• Managing multiple services without constant restarts
• Keeping frontend and backend in sync during development

• ✅ Built a working end-to-end upload + processing pipeline
• ✅ Clean UI with smooth user flow
• ✅ Proper separation of frontend, backend, and ML services
• ✅ System ready for full RAG integration

• Real-world AI systems are more about pipelines than models
• Separation of services = faster development
• Importance of containerization with Docker
• Embeddings + vector search are the backbone of modern AI apps
• Building systems > just writing code

• Improve performance, scaling, and deploy to web
• Turn it into a full AI document assistant
• Add authentication and user management
• Support more file types beyond PDF

Pull requests are welcome! For major changes, please open an issue first to discuss what you'd like to change.

This project is licensed under the MIT License.