
Volyx Lens
A privacy-focused desktop assistant for selected screen, microphone and meeting context. Only the context you attach reaches the provider.
Open to AI Engineer roles
Retrieval, context and evaluation. I build systems whose answers can be inspected, then ship the checks that prove it.
5 projects 12 upstream PRs 5 essays
Each project has a check that would fail if the output were wrong. Open the source, read the evaluation, or use the product.

A privacy-focused desktop assistant for selected screen, microphone and meeting context. Only the context you attach reaches the provider.

Local review analysis with citation validation and a labelled retrieval evaluation. Every answer can be traced back to a source.

Leakage-aware forecasting with chronological train, calibration and test windows. The test set is frozen before evaluation.
Not a demo with a polished screenshot. The context boundary, the citation check and the evaluation are all part of the work.
Attached context
Nothing is captured until you switch it on. The renderer is sandboxed with context isolation, and each provider gets only the context you selected.
LOCAL REVIEW
The retry limit moved into config.js. Two callers still read the old constant.
Local Review Intelligence checks every citation and gives an answer one bounded repair attempt when a source is missing or invalid.
View evaluation results (JSON)30-case retrieval benchmark
Retrieval is scored against a BM25 baseline. Forecasting is scored against a bookmaker-implied benchmark with chronological train, calibration and test windows and a frozen holdout.
Twelve pull requests accepted by other maintainers. Every row opens the accepted pull request, not a screenshot of one.
Open source
12
Agent frameworks, stream processing, data tooling and developer infrastructure.
Across 8 repositories
Essays on retrieval, agent memory, tool contracts and voice state.
Open to AI Engineer roles
I’m looking for AI Engineer or Applied AI Engineer work on Python services, retrieval and context systems, multimodal or voice workflows, and evaluation.