Software Engineer · Full-Stack Systems & AI
“Impeccable, on-time work — and generous about sharing knowledge with the team.”
A WhatsApp agent that recommends local businesses, in daily use by around 15 merchants. It came to me as a brittle no-code wrapper that broke if you looked at it sideways. I rebuilt it on a Python and LangChain stack I could actually steer — predictable, debuggable, and ready to grow past the demo it started as.
Read case study →An online store for high-value agricultural machinery — where one-click checkout makes no sense because nobody buys a harvester on impulse. I modeled checkout as an intention to purchase, with partial deposits and follow-through.
Read case study →Three insurance products, now live at a major bank, built on top of decades-old Java systems that weren’t going anywhere. Leading a team of four, I owned the seam between old and new — a backend-for-frontend that let a modern interface and a legacy mainframe talk to each other without either one dictating the terms.
Read case study →A legaltech platform for cap tables and data rooms, where a wrong click can become legal exposure. The hardest flows had grown into conditional rendering and boolean flags: half-submitted issuances, ambiguous signature states, screens nobody had explicitly designed.
System shape: explicit XState machines make every transition named, reviewable, and testable.
Architecture move: Module Federation split cap-table, data-room, and admin surfaces so teams could ship on their own clocks.
Outcome: the heaviest flows became diagrams the team could reason about instead of hidden state combinations.
A full-stack tool that listened to recorded calls and pulled structured, client-specific insight out of raw audio. I wired React to a Python backend, layered speech-to-text AI on top, and shaped its messy output into something a business could actually read. This was 2021 — AI in a product wasn’t the default yet; here it was the point.
Read case study →Therapeutic continuity for kids with a disability certificate — validated micro-activities delivered to families over WhatsApp. Winner, Startup Weekend Rosario ’25.
Open lab note →A curated craft marketplace paused after user interviews exposed the real bottleneck: not a better storefront, but distribution and demand.
Open lab note →A recruiting memory system that pivoted from Talent Rediscovery into a broader architecture for reliable AI-assisted expert work — provenance, durable memory, and human confirmation as the trust contract.
Open lab note →An idea for an app, feature or refactor — challenged before implementation begins.
Cache findings for difficult 'explore' phases in a research.md asset.
Hash out ideas in code for early feedback and reusable assets.
A markdown spec of the destination, with user stories and implementation notes.
Gherkin scenarios from the user stories — the behavior contract before any code.
Turn the PRD into individual tickets with blocking relationships.
Failing tests per ticket before implementation. Red → green → refactor.
Use AI-assisted implementation against ticket, test, and review constraints.
A QA plan for a human to verify the completed work.
The operating surface behind the process: the tools, system boundaries, and delivery practices that make the workflow shippable.