RAG-powered document intelligence — upload any PDF and chat with it using natural language. The system retrieves the most relevant sections and generates accurate, source-cited answers in seconds.
Built by Abdullah Shafique
AI Engineer · FastAPI · Groq · React
Most document tools make you search. This one makes you ask.
The PDF Knowledge Assistant lets you upload any PDF — a research paper, legal contract, company report, or policy document — and ask questions in plain English. It retrieves the most relevant sections using vector similarity search, feeds them to an LLM with full context, and returns a grounded, accurate answer with page-level citations.
No ctrl+F. No manual skimming. Just a question and an answer.
Try it live → pdf-knowledge-assistant.vercel.app
User uploads PDF
│
▼
[PDF Processor] ── pdfplumber extracts text page by page
│
▼
[Chunker] ── Text split into overlapping chunks (800 chars, 150 overlap)
│
▼
[Embedder] ── sentence-transformers converts chunks to 384-dim vectors
│
▼
[FAISS Store] ── Vectors indexed locally for cosine similarity search
│
▼
[RAG Engine] ── User query embedded → top-5 chunks retrieved
│
▼
[Groq LLM] ── Llama 3.1 generates grounded answer from context
│
▼
[React Frontend] ── Answer displayed with source citations + page numbers
- Conversational document search — ask questions in plain English, get cited answers
- Source citations — every answer shows which page and section it came from, with a relevance score
- Multi-document support — upload multiple PDFs and search across all or filter by one
- Multi-turn memory — rolling conversation history for follow-up questions
- Local vector store — FAISS runs entirely on-device, no cloud database needed
- Fast inference — Groq's Llama 3.1 8B delivers answers in under 3 seconds
- Clean dark UI — built with React, Vite, and Tailwind CSS
| Layer | Technology |
|---|---|
| Frontend | React 18, Vite, Tailwind CSS |
| Backend | FastAPI, Python 3.11 |
| LLM | Groq — llama-3.1-8b-instant |
| Embeddings | sentence-transformers/all-MiniLM-L6-v2 |
| Vector Store | FAISS (local, no cloud needed) |
| PDF Parsing | pdfplumber |
| Backend Deploy | Hugging Face Spaces (Docker) |
| Frontend Deploy | Vercel |
pdf-knowledge-assistant/
│
├── backend/
│ ├── app/
│ │ ├── main.py # FastAPI app + CORS + startup
│ │ ├── config.py # Pydantic settings from .env
│ │ ├── models.py # Request / response schemas
│ │ ├── routers/
│ │ │ ├── documents.py # Upload / list / delete PDFs
│ │ │ └── chat.py # RAG Q&A endpoint
│ │ └── services/
│ │ ├── pdf_processor.py # Extract + chunk PDFs
│ │ ├── embeddings.py # sentence-transformers (free, local)
│ │ ├── vector_store.py # FAISS upsert + similarity search
│ │ └── rag_engine.py # Retrieval + Groq generation
│ ├── requirements.txt
│ ├── Dockerfile
│ └── .env.example
│
├── frontend/
│ ├── src/
│ │ ├── components/
│ │ │ ├── Header.jsx # Top bar + branding
│ │ │ ├── Sidebar.jsx # Document upload + management
│ │ │ ├── ChatWindow.jsx # Message list + empty state
│ │ │ ├── ChatMessage.jsx # Markdown rendering + sources
│ │ │ ├── ChatInput.jsx # Auto-grow textarea
│ │ │ └── SourceCard.jsx # Expandable source citations
│ │ ├── hooks/
│ │ │ ├── useDocuments.js # Upload / delete state
│ │ │ └── useChat.js # Chat history + RAG calls
│ │ ├── api.js # Axios API client
│ │ └── App.jsx
│ └── package.json
│
└── assets/
├── banner.png
├── demo.png
└── architecture.png
- Python 3.11+
- Node.js 20+
- Free API key from Groq
# 1. Clone the repo
git clone https://github.com/aadi-abdullah/pdf-knowledge-assistant.git
cd pdf-knowledge-assistant
# 2. Backend setup
cd backend
python -m venv venv
venv\Scripts\activate # Windows
source venv/bin/activate # Mac / Linux
pip install -r requirements.txt
# 3. Create .env file
cp .env.example .env
# Add your GROQ_API_KEY inside .env
# 4. Frontend setup
cd ../frontend
npm install# .env
GROQ_API_KEY=gsk_...
GROQ_MODEL=llama-3.1-8b-instant
EMBEDDING_MODEL=all-MiniLM-L6-v2
EMBEDDING_DIMENSION=384
CHUNK_SIZE=800
CHUNK_OVERLAP=150
TOP_K_RESULTS=5# 5. Run backend (Terminal 1)
cd backend && uvicorn app.main:app --reload
# 6. Run frontend (Terminal 2)
cd frontend && npm run dev| Method | Endpoint | Description |
|---|---|---|
POST |
/api/v1/documents/upload |
Upload and index a PDF |
GET |
/api/v1/documents/ |
List all indexed documents |
DELETE |
/api/v1/documents/{doc_id} |
Remove a document |
POST |
/api/v1/chat/ |
Ask a question (RAG) |
GET |
/api/v1/chat/health |
Health check |
Interactive docs at http://localhost:8000/docs
| Variable | Description | Default |
|---|---|---|
GROQ_API_KEY |
Groq API key | — |
GROQ_MODEL |
LLM model name | llama-3.1-8b-instant |
EMBEDDING_MODEL |
Embedding model | all-MiniLM-L6-v2 |
EMBEDDING_DIMENSION |
Vector dimensions | 384 |
CHUNK_SIZE |
Chars per chunk | 800 |
CHUNK_OVERLAP |
Overlap between chunks | 150 |
TOP_K_RESULTS |
Chunks retrieved per query | 5 |
| Service | Free Tier | Link |
|---|---|---|
| Groq | Generous daily limits | console.groq.com |
- Streaming responses (SSE)
- Persistent document registry (PostgreSQL)
- Authentication (JWT)
- Re-ranking with cross-encoders
- Hybrid search (BM25 + vector)
I'm an AI Engineer with a background in design — I spent 6 years as a professional graphic designer before transitioning into software engineering. That background shapes how I build: I optimise for systems that are both technically sound and genuinely usable.
- 🎓 Software Engineering — Riphah International University (GPA 3.99 / 4.0, 2024–2028)
- 🏅 AI Agent Developer Specialization — Vanderbilt University
- 🏅 Microsoft Python Development Specialization
- 🏅 Adobe Graphic Designer Specialization
Currently open to AI engineering internships and junior roles.
→ LinkedIn · GitHub · abdullahshafique2019@gmail.com
MIT — free to use, modify, and distribute.