samudragupto/redrob-ranker

🏆 Redrob AI Candidate Ranking Engine

A deterministic, O(N) complexity candidate ranking system built for the Redrob Hackathon v4.

This engine evaluates 100,000 JSONL candidate profiles against a highly specific "Founding Engineer" job description in under 10 seconds on a standard CPU, entirely bypassing the 5-minute compute constraint.

🏗️ Architecture

Instead of relying on slow LLM API calls or easily-fooled dense embeddings, this system uses a 5-Dimension Rule-Based Heuristic Ranker modeled after actual engineering hiring rubrics:

1. 1.Core Technical (35%): Rewards deep production IR/RecSys experience. Penalizes pure CV/Robotics paths and AI keyword stuffers without engineering titles.
2. 2.Product Shipper DNA (25%): Rewards deployment terminology and pre-2022 retrieval experience. Penalizes pure academic backgrounds and LangChain-only portfolios without foundational ML depth.
3. 3.Career Stability (15%): Rewards 4-8 year sweet spots and internal promotion trajectory. Penalizes extreme job hoppers (<1.5yr avg) and pure-consulting backgrounds.
4. 4.Behavioral Signals (15%): Penalizes inactive candidates (relative to dynamic max active date) and low recruiter response rates.
5. 5.Logistics (10%): Enforces strict Tier-1 relocation and Pune/Noida preferences.

🛡️ Honeypot Defense

The dataset contains ~80 honeypots designed to trap semantic search approaches. This engine explicitly drops candidates before sorting using universal chronological impossibility checks.

🚀 Execution Instructions

Run Ranker:

python rank.py --candidates ./candidates.jsonl --out ./team_redrob.csv