Software · 2025– · TODO: status

Eval

Semantic search over a document library that works online or offline, at lower cost.

Role
TODO: role
When
April 2025–
Made with
Python, OpenAI API, sentence-transformers, PyTorch, scikit-learn

Problem

TODO: what documents, and whose problem.

What I did

I built an indexer that can switch between a hosted and a local embedding model, and cut redundant text before it gets embedded.

TODO: confirm one name: Eval or Textual Indexing Machine.

What changed

Retrieval cost dropped 40%. TODO: who uses it.

TODO: source and baseline for the retrieval-cost measurement. Confirm whether it is still active or Done.