Hi, I'm Tommy
Data Scientist & AI Engineer | Automation & ML systems
Data Scientist & AI Engineer based in Bogotá. I work at the intersection of ML models, LLM automation, and data systems at scale. Currently at Arbex (LATAM, formerly Kimberly-Clark) and building AI agents and automations with 2DATO, a project I run with two close friends.
Explore my technical notes and projects, or check the About section for more.
- ROLE
- Data Scientist & AI Engineer
- FOCUS
- Automation & ML systems
- LOC
- BOGOTÁ, CO
- LOG
- 012 ENTRIES
Featured
03-
Deepy — A local voice assistant for Windows, wired into the Claude Code CLI
How a hotkey-triggered voice assistant transcribes locally with Whisper, reasons through the Claude Code CLI with a local Ollama fallback, and talks back with Piper TTS — with almost nothing leaving the machine.
-
Velora — A supply-chain calculator suite with a privacy-first AI layer
Architecture behind a bilingual suite of free supply-chain calculators where an AI layer explains every result like a consultant would — without ever seeing the user's raw data.
-
Hi, I'm Tommy
Data Scientist building the path towards AI Engineer.
Recent Posts
04-
Deepy 2.0 — Rebuilding a personal assistant around one Postgres, a WhatsApp core and a private dashboard
Why I threw away the Google Sheets pipeline behind my voice assistant and rebuilt Deepy as a WhatsApp + Postgres core with a single-user Next.js dashboard that shares the same database.
-
Neo Tamagotchi — A pocket pet for my dogs, in C on an RP2350
A handmade tamagotchi with my family's dogs in pixel art: portable C11 game logic shared between a PC simulator and a microcontroller, two clocks that survive sleep, and an angel dog who never needs care.
-
Kana Practice — Learning to read hiragana by phonetically adapting Spanish
A static web app where Spanish words become kana by sound, you transcribe them back to romaji, and an edit-distance alignment grades every kana — no framework, no build step.
-
Gestorio — An in-house PMO with an MCP server so Claude Code can pick up tasks
How I built the internal PMO of 2DATO — tasks, notes, client portal and weekly updates — and exposed it through an MCP server where AI agents can move tasks but can never publish to clients without a human-approved confirmation code.