I build production systems where a language model does the perceiving and deterministic code does the deciding. Python · FastAPI · Go · PostgreSQL · AWS.

Voxa · tryvoxa.com Voice-to-spreadsheet capture: you upload an Excel template, Voxa reads its schema, transcribes what you narrate and fills the row. FastAPI · React · PostgreSQL on AWS, provisioned with Terraform. Releases carry no long-lived credentials and no inbound SSH — GitHub OIDC into AWS, multi-arch images to GHCR, rollout through SSM. 600+ tests that run with no network and no API key.

voxa-core The engine underneath, extracted once a second product showed how much of the first was domain-agnostic: transcription, the LLM call with retry and error translation, tolerant JSON parsing, the data contracts. Products consume it through a single prompt-builder seam, so no domain leaks into the generic path.

PlayPro Stats · playprosystems.com Voice play-by-play statistics for American football, built for a paying client. The model extracts only what was narrated — an unmentioned field stays absent, never a courtesy zero — while a deterministic rules engine computes downs, possession, turnovers, yard-line math and scoring. An eval harness scores per-field accuracy against a golden narration set. Private repo; the engine above is public.

How I work. Test-first, and the tests run offline by design. Every non-trivial decision is written down as an ADR in the repo, with the alternatives I rejected and what the choice costs — 19 in Voxa, 28 in playPro Stats.

The full portfolio, with the trade-offs spelled out → cvegaz.github.io