Traceable answers
Return sources alongside generated responses so readers can inspect the underlying material.
Teaching tool · Applied AI / Full-stack · 2025
CV Screener
A teaching reference for building a document-search system: PDF ingestion, hybrid retrieval and answers linked to their sources.
The problem
Make a collection of PDF resumes searchable through questions, while keeping the source material available alongside each answer.
The approach
Separate document ingestion from the serving API. Combine vector retrieval with BM25, store documents and embeddings in PostgreSQL, and expose the same query interface to the web client. Cache responses and instrument the API with Prometheus.
What the project delivers
A runnable end-to-end project with PDF ingestion, hybrid retrieval, cited answers and a Next.js interface. Docker Compose coordinates the services, and the repository includes API and embedder test commands.
Design decisions
Separated services
Keep ingestion, storage, retrieval and the interface independently understandable.
Operational visibility
Include caching and metrics in the serving path.