Adaptive RAG — Frontend

Setup

1. Install dependencies

npm install

2. Configure the API URL

Create .env.local:

NEXT_PUBLIC_API_URL=http://localhost:8000

3. Run the dev server

npm run dev

Open http://localhost:3000.

---

Features

Normal Mode (dev toggle OFF)

• Upload a PDF
• Select no pruning strategy (default)
• Ask a question, get an answer with source pages

Dev Mode (click </> dev in navbar)

Exposes the full backend pipeline:

1. Pruning Strategy Selector — choose between none, cosine, cosine_whitened, kmeans, mmr
2. Pipeline Inspector — step-by-step breakdown of ingestion → chunking → embedding → pruning → storage
3. Pruning Report — per-chunk scores, threshold, retention rate, score distribution
4. Embedding Visualizer — 768D vector heat map (sampled) per chunk
5. Query Parameters — temperature slider, max tokens, MaxSim re-ranking toggle

Strategy reset

Switching pruning strategies automatically calls /reset before re-uploading, so stale vectors from a previous strategy don't pollute the new run.

---

Backend Setup

This frontend requires the RAG backend to be running. See the backend repo for full setup instructions. In short:

# Install Python dependencies
pip install fastapi uvicorn pymupdf spacy sentence-transformers transformers
pip install bitsandbytes accelerate psycopg2-binary pgvector python-dotenv
python -m spacy download en_core_web_sm

# Configure .env with your DB credentials and HuggingFace token
# Then start the server
uvicorn main:app --reload --host 0.0.0.0 --port 8000

---

API endpoints used

Method	Endpoint	Purpose
POST	/upload?pruning_strategy=	Upload PDF, starts background processing
GET	/status	Poll processing state (polled every 1.5s)
GET	/pruning-report	Full per-chunk pruning stats
POST	/query	Ask a question, get answer + sources
GET	/reset	Clear pgvector + in-memory state