Using AI to leverage impact.

I research and develop AI tools and platforms to enable, enhance, and inform public policies and social projects.

Prototypes

I create minimal models of websites and tools to test ideas.

Platforms

I create complete platforms for end users.

Impact research

I collaborate on the academic research that grounds projects.

Leo H. M. Arruda speaking into a microphone

Selected projects

View all →

AI Model calibration and task optimization

In development

BenchTask

A platform for comparing AI models through tasks and standardized suites for prompt optimization and model choice.

(more info)

Law project risk analysis

White paper

The Capability Blind Spot in Brazil’s AI Bill

An analysis of risks from agentic AI systems in the context of Brazilian Bill 2,338/2023.

Background and experience

I work with AI for evidence synthesis and build tools for impact projects and public policy.

I have a background in scientific research, including an MSc in Psychobiology at UNIFESP. With Instituto Veredas, I co-authored a review published in PLOS ONE commisioned by WHO on integrating evidence into Theories of Change.

Four years as a Python developer in the financial and real-estate sectors gave me hands-on experience with data pipelines, APIs, and automated information extraction from legal documents. I now apply that experience with AI through document retrieval and model evaluation in tools such as Assist3R and BenchTask.

In the impact ecosystem, I worked with Effective Altruism Brazil and Condor Initiative, in strategy and evaluation.