RAGLAB

Configurable Pipeline & Multi-Metric Evaluation Platform

PostgreSQL
Groq LLM

1. Upload Document

PDF parser with automatic page & chunk indexing

Click to browse or drop PDF here Maximum upload size: 10 MB No file chosen
FAISS index per-session
Upload a PDF to start retrieval and evaluation.

2. Configure Pipeline

Tune chunking, embedding models, top-K, and CrossEncoder reranking

500
Range: 100 – 2000 characters
100
Must be strictly less than chunk size
Changing model re-embeds current document
5
Chunks supplied as context to LLM (1–20)
CrossEncoder Reranking ms-marco-MiniLM-L-6-v2

Two-stage pipeline: FAISS retrieves 2x candidates → CrossEncoder reranks to Top-K.

Changes apply automatically • stored with versioning in PostgreSQL

3. Ask Question

Retrieval • Groq Llama 3.3 70B • Multi-Metric Evaluation • Latency Tracking

Answers generated strictly from retrieved context. If not present in text, system reports unavailable. Press Enter to submit
Upload a PDF and run a query to inspect the generated answer, raw multi-metric evaluation scores, and latency breakdown.

5. Query & Config History

Persisted in PostgreSQL query_logs

No query history logged in PostgreSQL yet. Run your first query above!

6. Download Your Pipeline

Select components • Automatic dependency resolution • Clean Python code

Export Format:
Zero hard-coded credentials