Talent Intelligence AI
Built for the RedRob AI × Hack2skill India Runs Hackathon in the Intelligent Candidate Discovery track.
Talent Intelligence AI
Offline candidate ranking system.
Context
An offline candidate ranking pipeline designed to evaluate resumes against job descriptions systematically. It finds actual evidence of skills rather than relying on standard keyword overlap.
Applicant tracking systems often filter out qualified candidates because they lack exact keyword matches, while prioritizing candidates who stuff their resumes with buzzwords.
Execution
- 01Official 100K candidate dataset processing
- 021M synthetic candidate stress testing
- 03CPU-only deterministic ranking
- ADesigned the evidence scoring logic to move beyond basic keyword matching.
- BImplemented the offline evaluation pipeline.
- CAnalyzed candidate signals and calibrated the ranking system.
Architecture Flow
Deep Dive
Engineering Decisions
We strictly opted for a CPU-only deterministic ranking approach instead of an LLM system. This decision was driven by the need for explainable, reproducible, and highly performant evaluations at scale.
Challenges Overcome
Calibrating the ranking weights was difficult. We had to ensure the system didn't over-index on isolated keywords and required supporting evidence (e.g., tenure, project context) to assign high scores.