Problem
Recruiters lose hours to manual resume screening, and the judgement that comes out of it is inconsistent between reviewers and hard to defend.
Solution
A multi-stage Gemini pipeline: candidates arrive by form, CSV, Excel or PDF, are scored in parallel batches, then globally re-ranked so scores stay comparable across batches. Every score carries the reasoning behind it.
My Role
Built with team Innovatex (Diane Uwamariya, Bugingo Innocent, Nizeyimana Elie). My part: the Next.js and TypeScript frontend on Vercel, the Node service on Render, the MongoDB models, the Gemini screening pipeline and the per-candidate email system.
Challenges
Keeping scores fair across batches, and keeping the model honest: the pipeline flags gender language, age indicators, institution prestige and name bias inside its own output, and surfaces unverified skills and employment gaps as explicit risks.
Impact
Screening that took hours returns a ranked shortlist in minutes, with strengths, gaps and interview questions per candidate — and a reasoning log a recruiter can audit rather than take on trust. The team placed 4th at the Umurava AI Hackathon, recognised for the best HR-industry context and user experience.
Architecture
Next.js on VercelNode + TypeScript on RenderMongoDBGemini screening pipelineBatch scoring + global re-rank