The challenge
Resume search often fails when recruiter language and candidate wording differ, especially for synonyms, seniority and related skills.
RAG / NLP
Resume Analyzer with RAG and Reranking
A retrieval-augmented resume search and analysis system using section-aware chunking, embeddings, metadata filters and reranking.
Role
AI / ML Engineer
Context
Independent Build
Period
2025–2026
What I built
Built a RAG pipeline that chunks resumes by semantic sections, stores embeddings with structured metadata, filters by role and seniority, and reranks retrieved candidates with a cross-encoder. Added citation-style evidence so matches could be verified against the source resume.
Architecture
- Section-aware resume chunking
- OpenAI / BGE embeddings
- Pinecone or pgvector retrieval
- Role, skills and seniority metadata filters
- Cross-encoder reranking and source citations
Outcomes
- Evaluated with precision@k and MRR
- Improved synonym matching with a domain dictionary plus reranking
- More explainable candidate matches through source evidence
Technology
PythonRAGBGEOpenAI EmbeddingsPineconepgvectorCross-Encoder