lara@sf:~$ whoami
Lara
Barijhoff_
I build and ship AI products. Moving from commercial AI into safety.
lara@sf:~$ cat focus.md
I've spent the last two years building commercial AI products: no-code agents shipped to real businesses, then a self-hosted agent ecosystem I use every day. I'm moving that work into AI safety as a builder, hands-on with systems rather than research.
I'm in the exploratory part of that move, deliberately. I went to EAG Berkeley to meet people actually doing this work and came out with a much clearer map of my next steps. Right now I'm building context: forming my own view on how fast capabilities are moving, which problems that makes most pressing, and which approaches I actually find convincing. That could land as a role on a team, or as founding something of my own.
// about
I'm an engineer who ships. What I'm good at is taking something from a vague idea to a working thing people can actually use, fast. I'd rather have something running and imperfect in front of a real user this week than a clean plan for next month.
I graduated at the end of 2025: Digital Technology at Universidad Torcuato Di Tella, with a minor in Innovation & Entrepreneurship and a semester at ESSEC in France. Being this early is part of why I'm exploring. I've shipped enough to know what I'm good at, and I'm still working out what I want to point it at.
Right now I'm in San Francisco, running a self-hosted ecosystem of AI agents and shipping small products out of it. Two of them are below, both deployed and in daily use.
What I'm looking for next is where to apply that in AI safety: someplace close to where agents meet real users. The thread I'm pulling on today is evals for mental-health-relevant AI interactions, out of Fifty Years' AI RFS.
// experience
AI Validator
Evaluated model outputs at scale for a pre-seed startup, correcting the cases the pipeline got wrong.
- ▹Manually validated AI-extracted profiles for 2,746 founders across 1,570 companies.
- ▹Identified and corrected outputs falling below the 75% accuracy threshold, improving match reliability across the full dataset.
- ▹Hands-on with model failure modes and edge cases: where outputs break, and why.
// building now
AI Agent Ecosystem
Mar 2026 — Present
A self-hosted suite of specialized AI assistants running on OpenClaw with MCP server integrations for automated reminders, running training plans, and research workflows. Configured with Claude Code and Codex. The two products below came out of it.
shop-itinerary
2026 · deployed
Drop in a screenshot of a product page, a checkout, or a wishlist, and Claude vision extracts the items into a structured shopping itinerary for trips abroad. Full-stack, deployed, in daily use.
// shipped before
Nubiq
Mar 2025 — Dec 2025 · Co-founder
Co-founded an AI startup building no-code agents for SMEs. Ran operations and built the front-end. Shipped 3 agents (stock checker, supplier finder, FAQ chatbot) and completed 5 pilots with real businesses.
Afrodita Accessories
Jan 2021 — Jul 2022 · Co-founder at 17
Co-founded a gemstone accessories venture at 17, running logistics and influencer marketing across Argentina. Grew revenue 150% in 3 months, closed 2 influencer partnerships, 500+ sales.
// education
Universidad Torcuato Di Tella
Software engineering, data science, systems design, digital business strategy, and applied technology. Strong skills in Python, C++, and full-stack development.
- ▹Exchange semester at ESSEC Business School, France (2024).
- ▹Coursework: Machine Learning, Databases, Full-Stack, Digital Strategy, Neuroscience & Design.
Escuela Argentina Modelo
Strong international academic profile with multiple language certifications.
- ▹CAE C1 Advanced English (2021) · FCE B2 (2018).
- ▹IGCSE English as a First Language & History · Advanced Portuguese.
// skills
// contact
I'm open to conversations about AI safety, engineering, and what I could build with your team.