About This Project
Overview
AI Resume Screener replaces the 6-second resume skim with a structured Gemini 2.5 Flash evaluation: every PDF resume is scored 0–100 against the job description, summarized, and ranked, with strengths and weaknesses extracted for the admin view. Once a candidate is screened, admins can one-click generate a tone-matched follow-up email (next-step, offer, or decline) that references real strengths without ever leaking the score. FastAPI and SQLModel persist everything to SQLite so results survive restarts.
Motivation
Manual resume screening is brutally repetitive and inconsistent, and I wanted to see if a well-structured LLM prompt could do a fairer, faster first pass than a tired recruiter skimming for six seconds a resume.
Lessons Learned
The hardest part wasn't the scoring — it was designing the email-drafting prompts so they'd reference real strengths without ever leaking the underlying score or admitting AI was involved. That constraint forced me to get much more deliberate about prompt boundaries.