Senior AI Research Engineer · Mercado Libre
Jairo Torregrosa
I came up through evaluation: first models, then agent trajectories. Now I build the agents I used to grade.
available bogotá, colombia utc-5
About Me
I am a Senior AI Research Engineer at Mercado Libre, based in Colombia. I started in evaluation, picking cost-effective models with Inspect AI and later grading agent trajectories. Today I build agents on the Applied Agents team. Day to day I design agent harnesses, write evals, and build the tools agents need to do real work.
Before moving into GenAI and agentic systems, I worked on production machine learning, recommender systems, ranking, and computer vision. That background still shapes how I think about AI systems: define the task clearly, measure behavior, inspect failures, and keep the system grounded in real constraints.
I also serve as a Claude Community Ambassador in Bogotá, helping builders, founders, and technical teams put agentic engineering into practice.
From evals to agents
Evaluating models
The goal was never the best model. It was the cheapest one that still cleared the bar for each task. At Mercado Libre I built eval suites in Inspect AI, scored candidates against task rubrics, and ranked them on quality per dollar.
Evaluating agents
Same discipline, harder target. A single answer became a multi-step trajectory, so the question changed from "did it answer well?" to "did it take the right path?" Did it act correctly, recover from its mistakes, stay in policy? Grading the trajectory, not just the outcome.
Building agentsNow
Now I build what I used to grade, working on the personal agents platform with the Applied Agents team at Mercado Libre. The evaluator habits came with me: I'd rather measure an agent than trust its demo.
Experience
Mercado Libre
- Senior AI Research EngineerCurrent
- Senior Data Scientist
- Data Scientist, Applied Machine Learning
Earlier roles
Data Scientist
Tul
Jan 2022 - Dec 2022
Data Scientist
Asesoftware
Jan 2021 - Dec 2021
Software Developer
BTG Pactual Colombia
Jan 2020 - Dec 2020
Public Work
jaiskills
Open source · Claude Code plugins
Claude Code skills and plugins, including Insistir, a multi-agent workflow for implementation, review, revision, and judging.
Python · Claude Code · Multi-agent
SmolNews
Hugging Face Space
A small agent that fetches the news and the local time for any place you ask about. Built on the agents template and running as a live Space.
Python · Agents · Hugging Face
Taste with Your Eyes
Public app
A public app for turning restaurant menus into visual digital experiences.
Next.js · AI
Community
Claude Community Ambassador in Bogotá, focused on Agentic Engineering.
I help organize hands-on sessions for builders, founders, solopreneurs, and technical teams working with Claude and agentic AI. We build on real products, under real constraints, and share what we learn.
Top Skills
AI Agents · Agentic Engineering · Harness Engineering · Agent & LLM Evaluation · Claude Code · Tool Use · Python · Machine Learning · Recommender Systems · Computer Vision
Education
Pontificia Universidad Javeriana
Bachelor's degree, Electronics Engineering
2014 - 2020
- Honorable mention thesis
Certifications
Building Agentic AI Applications with LLMs
NVIDIA
AI Agents Fundamentals
Hugging Face
Publications
Comparing state-of-the-art methods of detection and tracking people on security cameras video
IEEE - STSIVA 2019
Show publication →Tertiary roads segmentation using deep learning from satellite imagery
Thesis · Pontificia Universidad Javeriana, 2020
Show publication →