Job application pipeline: personal tooling

The tool behind the tool: a Python backend built to solve my own actual problem, generating evidence-based cover letters and CVs without either making claims up or spending hours re-reading old documents for the right proof point every time.

A SQLite database holds structured, ingested evidence pulled from the real archive: service letters, project notes, cover letters I'd already written, arranged so a claim can be traced back to its actual source rather than asserted from memory. Retrieval is hybrid: sentence-transformers embeddings for semantic search, combined with keyword matching, so a request for "evidence of leading a migration" finds the Kubernetes work at TrueStory even if that project's own notes never used the word "migration" verbatim. Results get ranked and surfaced as proof points that then drive the actual generation step.

This project and the career graph you're looking at right now, and its Kuzu-based Cypher extraction pipeline, are the same lineage of tooling, evidence-sourced generation applied first to documents, then to a queryable graph. Which makes this one slightly recursive: the tool that helped source the evidence behind this graph is itself one of the graph's own entries.