I build conversational applications with Python, FastAPI and React, with a focus on the integration work between model APIs, business functions and browser clients.
Vexium Voice is a bilingual Spanish/English receptionist reference app using Qwen and Next.js. It includes text and voice paths, appointment workflows and an event dashboard.
- Architecture and operating boundaries
- Automated checks: Python regression tests, browser smoke tests, lint and builds.
- Contributor setup with locked dependencies and mocked providers.
- Guía en español
The project is a reference application under development. Its documentation identifies simulated integrations and the work needed before production use.
I am starting to contribute reproducible bug reports, regression tests and focused fixes to the tools these applications depend on. Current work:
- Starlette: IPv6 URL replacement report and tested patch, awaiting maintainer feedback.
- pytest: reproduced a fixture teardown regression in an existing PR, review submitted.
- aiohttp: expire cookies at the Unix epoch, draft fix with unit and HTTP regression tests.
- Tenacity: flatten asynchronous retry predicate chains, open PR with recursion and short-circuit regression tests; CI passing.
- gpxpy: treat zero elevation as valid track data, open PR covering movement, elevation availability and smoothing.
- Babel: ignore quoted text when selecting date and time fields, open PR with six regression cases.
- python-semver: reject malformed version strings, open PR enforcing ASCII digits and complete input validation.
- pathspec: allow subtree traversal from filesystem roots, open PR covering filesystem roots and Windows path casing.
I use AI coding tools with attribution. Public tests, implementation notes and discussion links document the work and its current review status.