CoolFace
Apppublic

syedmohsin7/redrob-ranker

sourceHugging Faceupdated 3mo agoView on Hugging Face
1likes
App README

Sentinel-AI — Team cache_Q

High-throughput candidate ranking pipeline for the RedRob AI Talent Search Challenge.

Overview

This repository contains a production-ready, zero-network-dependency Python pipeline that:

  1. 1.Streams 100,000 candidate profiles from a JSON Lines file
  2. 2.Filters adversarial honeypot profiles using deterministic arithmetic checks
  3. 3.Scores candidates with a composite technical + behavioral ranking matrix
  4. 4.Exports a perfectly formatted, monotonically non-increasing CSV of the top 100 candidates

All processing completes on a standard CPU within a 5-minute wall-clock budget.

Quick Start

1. Install Dependencies

bash
pip install -r requirements.txt

2. Download Model Cache (For Offline Execution)

bash
python download_model.py

3. Run the Pipeline

bash
python rank.py --candidates ./candidates.jsonl --out ./team_cache_Q.csv
ArgumentDefaultDescription
--candidatescandidates.jsonlPath to JSONL candidate dataset
--outteam_cache_Q.csvOutput CSV file path

4. Validate Output

The pipeline automatically invokes validate_submission.py at the end of execution. You can also run it independently:

bash
python -c "from validate_submission import validate_submission; validate_submission('team_cache_Q.csv')"

Output Format

The output CSV (team_cache_Q.csv) contains exactly 100 rows with the following columns:

ColumnTypeDescription
candidate_idstringUnique candidate identifier
rankintegerSequential rank from 1 (best) to 100
scorefloatComposite score, monotonically non-increasing
reasoningstringTiered explanation (Elite / Strong Fit / Borderline)

Scores are sorted descending; ties are broken by candidate_id in ascending lexicographic order.

Pipeline Architecture

candidates.jsonl
        │
        ▼
 ┌─────────────────────────────────────┐
 │  STAGE 1: Gated Streaming Filter   │
 │  • Honeypot arithmetic checker      │
 │  • Title adjacency guard            │
 │  • Industry verification layer      │
 └──────────────┬──────────────────────┘
                │ surviving candidates
                ▼
 ┌─────────────────────────────────────┐
 │  STAGE 2: Technical Scoring        │
 │  • Experience curve mapping         │
 │  • Core skill intersection          │
 │  • Execution context scoring        │
 └──────────────┬──────────────────────┘
                │ S_tech
                ▼
 ┌─────────────────────────────────────┐
 │  STAGE 3: Behavioral Calibration   │
 │  • Temporal inactivity decay        │
 │  • Response rate index              │
 │  • Notice period alignment          │
 │  • Geographic weighting             │
 │  • GitHub velocity scoring          │
 │  • Assessment validation            │
 └──────────────┬──────────────────────┘
                │ M_behavior
                ▼
 ┌─────────────────────────────────────┐
 │  STAGE 4: Aggregation & Reasoning  │
 │  Score_final = S_tech × M_behavior  │
 │  Sort → Top 100 → Rank → Explain   │
 └──────────────┬──────────────────────┘
                │
                ▼
 ┌─────────────────────────────────────┐
 │  STAGE 5: Validation & Export      │
 │  • Format check                     │
 │  • Monotonicity assertion           │
 │  • validate_submission() gate       │
 └─────────────────────────────────────┘
                │
                ▼
         team_cache_Q.csv

Repository Structure

rehob.ai/
├── rank.py                              # Main pipeline script
├── app.py                               # Gradio UI for HuggingFace Spaces
├── requirements.txt                     # Pinned Python dependencies
├── submission_metadata.yaml             # Challenge metadata declarations
├── validate_submission.py               # Provided submission validator
├── candidates.jsonl                     # Input dataset (not committed)
├── team_cache_Q.csv                     # Output submission (generated)
└── README.md                            # This file

Environment

  • —Python: 3.10+
  • —OS: Any (tested on Windows 11, Ubuntu 22.04)
  • —Network: Not required — fully offline execution
  • —Hardware: Standard CPU (no GPU needed)

Team

Team ID: cache_Q Challenge: RedRob AI Talent Search — Candidate Ranking Pipeline

License

This project is submitted as original work for the RedRob AI Talent Search Challenge.