Afuu-coder/redrob-ai-ranker

🧠 Redrob AI Candidate Ranker (Sandbox)

Interactive demo of the Intelligent Candidate Discovery & Ranking System for the Redrob Data & AI Challenge (Track 01).

What this sandbox does

Judges can upload a candidate sample (≤100 candidates, JSON or JSONL) and the app ranks them against the Senior AI Engineer (Founding Team) JD in seconds, producing a ranked shortlist with per-candidate reasoning.

This sandbox runs the BM25-fallback ranking path (no pre-computed embeddings required), so it stays lightweight on HuggingFace Spaces' free tier. The full semantic pipeline (sentence-transformer embeddings + TF-IDF hybrid retrieval) lives in the GitHub repo and is what produces the submitted submission.csv on the full 100K pool.

Features

• —📂 Upload a candidate JSON/JSONL sample (or use the built-in 50-candidate demo)
• —🏆 Ranked shortlist with scores and per-candidate reasoning
• —⚠️ Honeypot & JD-disqualifier flags highlighted
• —🔎 Expand any candidate to see the scoring breakdown across 7 dimensions

Files

• —app.py — Streamlit app (imports the ranker from rank.py)
• —rank.py — the 6-stage ranking engine (BM25 path active in this sandbox)
• —sample_candidates.json — 50-candidate demo dataset
• —requirements.txt — pinned dependencies

Full repo

The complete implementation (semantic embeddings, disqualifier gate, eval harness, reasoning generator) is at: https://github.com/Afuu-coder/redrob-ai-ranker