JorgeCabello
Product Engineer — Buenos Aires
Six years at startups, from seed rounds to Series B.
I build products with taste.
Terminal
About
Software engineer with 8+ years building scalable web applications. Went from frontend to full stack to AI engineering.
Joined as the second engineer at a startup and helped grow it to 30 people through Series B. Always early, always building the core architecture. Currently at Cometa (Series A, $12M), working alongside the team on AI projects — vector search, RAG, and multi-agent systems.
I treat AI as infrastructure. I build with Claude Code and Codex as daily collaborators, creating custom tools, multi-agent systems, and automated workflows to ship faster. I still care deeply about craft and how things feel, which is why I rebuild things like this portfolio just to experiment.
Experience
Experiments
Zas
macOS menu bar app that runs speech-to-text locally on Apple Silicon. No cloud, no waiting. Press a key, talk, it types. 25 languages, filler word removal, runs entirely offline.
Voiz
iOS journaling app where you talk instead of type. Local transcription, emotion tracking over time, AI-generated insights about your patterns.
ClaudioMac
Native macOS app for managing Claude Code sessions visually. Liquid Glass UI on macOS Tahoe, embedded terminal via SwiftTerm, session state detection. Also has a Telegram bot that lets me control sessions from my phone — stream output, send commands, manage worktrees.
Hand
Personal AI fleet manager. Decomposes tasks, dispatches parallel Claude Code sessions, and runs a fleet of autonomous bots — each with its own persona, interface (Telegram, WhatsApp, CLI), and persistent memory.
Voice Agents
Voice agent platform built on LiveKit + Twilio SIP. Inbound customer support with pluggable LLM backends, plus an outbound agent that calls with a mission, reports findings, and escalates decisions to a human mid-call.
Stack & Tools
Workflow Automations
Multi-model code review. Sends my current context to Codex for an independent analysis, then compares both perspectives. Catches blind spots neither model finds alone.
End-to-end production error resolution. Finds the error in Sentry, traces it to the code, applies a fix, opens a PR, and monitors the deploy for regressions.
Semantic search over 7,100+ files of past conversations and notes. Gives every new session full context from every previous one.
Browser automation that logs into my bank, pulls transactions, categorizes spending, and updates my financial tracking. Full Playwright pipeline.
Navigates Argentina's tax portal to generate export invoices. Form filling, validation, submission. All through browser control.
Natural language control of Philips Hue lights. Mood presets, room awareness, bilingual commands. Hardware integration through Claude Code.