I am a Research Scientist with a solid background in Software Engineering and an unconventional path to Research.

More recently, I have been involved with the next generation of LLM-based Agents, focusing on pretraining of Foundation Models with focus on Trading for Quantitative Finance as well as in post-training RL finetuning for Reasoning.

Before that, I worked with RL finetuning for LLMs (RLHF / RLAIF), with focus on AI Alignment in the OpenAI core models team at Microsoft in the Azure AI Research Team.

I received my Master’s Degree in Image, Vision and Machine Learning in a double degree program between ITA (Brazil) and École Polytechnique (France).

In previous teams at Microsoft, I worked with Knowledge and Language group and in Microsoft Research (MSR) as a Research Engineer. And in the Cloud Efficiency group, I contributed to several “Self-driving datacenter” scenarios through the Resource Central platform (ML prediction-serving System for Azure).

Research

My main Research agenda in that last years has been on Foundation Models for AGI (Artificial General Intelligence) with focus on Autonomy and Agency and can be summarized in three pillars:

• Cognitive Architectures (LLM + Prompt-Engine + Memory-Engine)
• Digital AGI (Skill Learning / Decision making / Tool Usage)
• Embodied Intelligence / Robot Learning

While I am searching for a path towards funding for my AGI Research, I create profitable solutions in real-world optimal control applications spanning from autonomous systems to data center energy optimization. In particular, currently I am involved on the adaptation / finetuning of Large Language Models with human inductive bias in the loop (AI alignment) and Reasoning for Quantitative Finance.

News and Updates

• 2023.10: I will be in ICCV presenting HoloAssist. Stop by to chat!

• 2024.12: I will be in NeurIPS. Let’s meet!

Contact

If you are interested in Cognitive Architectures augmenting LLMs with a Pre-frontal Cortex and External Memory with focus on Multi-modality, Embodiment and Skill Learning, please ping me.