AI Engineer & Builder
Less narration, more shipped AI.
I design and ship production AI systems — LLM-powered apps, RAG pipelines, multi-agent architectures, custom MCP integrations, evaluation frameworks. Currently leading AI development for a US-based biotech client; two decades building software and companies behind the judgment.
How I work
Production AI architecture
Designing the system around the model: retrieval, tool use, orchestration, state and memory, security boundaries, performance trade-offs. From single LLM-in-the-loop apps up through multi-agent systems — two decades of engineering applied to the new substrate.
Specification-driven development
A written spec becomes a working AI system rapidly and reproducibly. The spec — not the prompt, not the code — is the artifact. Compresses the gap between idea and shipped behavior, and gives the team something concrete to review.
Evaluation frameworks
Public benchmarks plus custom internal scenario suites — the holdout set the system never sees during development. Improvement gets measured, not declared. The right unit of test depends on the layer; both regression and exploratory evals belong in the pipeline.
AI-native delivery
I direct coding agents (Claude Code on Opus) as the implementation layer while owning architecture and quality. Operating at Level 4–5 of agentic engineering — the agents write the code, I write and review the specs.
Selected work
Production AI systems I've built and shipped. Detailed case studies land in the next iteration.
2025 — present
Astrolinkers — programmatic astrology API
Production-grade API for Western and Vedic natal charts, talent profiles, compatibility, interpretations, async PDF reports, public docs, SDKs, and live demo. Built solo; currently running in beta.
Python · FastAPI · PostgreSQL · Redis · RabbitMQ · Swiss Ephemeris
2025 — present
Alwenna — a personified AI astrologer
A consumer astrology product on my Astrolinkers API: a personified AI astrologer who reads your real chart and explains it plainly — flagship: compatibility. The LLM is strictly grounded; it only rephrases computed facts, never invents them. In beta.
TypeScript · Next.js · FastAPI · Python · Astrolinkers API
Mar 2025 — present
Scientific Research Agents — AI Architecture
Designing and leading the implementation of autonomous AI systems for scientific research automation at a US-based biotech client. Custom MCP servers, RAG pipelines over scientific corpora, and a held-out evaluation suite.
Python · FastAPI · Claude Opus · MCP · pgvector
2025 — present
Algodesks — algorithmic trading optimization
Crypto futures platform for strategy research, parameter optimization, backtesting, and Bybit paper trading. Optimization jobs run on Fly.io workers; no ML in the current system.
Python · FastAPI · Next.js · Fly.io · Bybit
2026 — present
Tributo — tax clarity for the self-employed
A mobile-first tax calendar for the self-employed in Uruguay, then LATAM: every DGI and BPS obligation with dates and amounts, and a rule never to invent a figure. In beta; built solo by directing AI agents.
TypeScript · Next.js · tRPC · PostgreSQL · Trigger.dev
AI Architecture Scorecard
Is your AI system production-ready?
Take the AI Architecture Scorecard — a short, opinionated self-assessment across six dimensions, with a prioritized read on what to fix first. About three minutes, no signup.
Take the scorecard →Let's build something that lasts past week two.
If you're putting AI into a production system — an LLM-powered feature, a RAG pipeline, an agent — and want a second pair of eyes, or if you're hiring for a senior AI architecture role — talk to me.
Get in touch