Problem

Job seekers struggle to pass Applicant Tracking Systems (ATS) without knowing which keywords, skills, or formatting issues cause their resumes to be filtered out before a human reads them.

Solution

Built a production-grade resume analysis platform that processes PDF/DOCX uploads through a multi-stage NLP pipeline: keyword extraction (KeyBERT), skill detection (spaCy NER), semantic similarity (sentence-transformers), and formatting analysis — producing a weighted ATS score out of 100.

Integrated Groq's Llama 3.3 70B for AI-generated improvement suggestions and bullet point rewrites. Results are persisted in Supabase with shareable links and per-user history.

Impact

• End-to-end analysis in under 15 seconds including LLM suggestions
• Shareable result links with persistent history for signed-in users
• PDF export of full analysis report
• Zero infrastructure cost using HuggingFace Spaces + Vercel + Supabase free tiers

Tech Stack

React 18, Vite, Tailwind CSS, Recharts, FastAPI, spaCy, KeyBERT, sentence-transformers, Groq API (Llama 3.3 70B), Supabase (PostgreSQL + Auth), HuggingFace Spaces, Vercel.