Aaron Tamte
AI/ML Engineer
I build state of the art AI agent systems. Currently building SheetOS and completing a Master's in Artificial Intelligence at Northeastern University.
Blog
Training a 4B Model to Search by Writing Code
Instead of giving a model a fixed retrieval tool, we trained it to write and execute its own Python search strategies. 3.7x improvement over the untrained baseline.
Fine-Tuning LLMs for Multi-Agent Collaboration
Can collaboration be trained into a language model? We built a benchmark of 7 coordination tasks and fine-tuned LLaMA-3 8B to match a model 9x its size.
Building a RAG System That Helps AI Agents Write Better Code
AI coding agents struggle with complex data science tasks when they don't have relevant examples. We built a retrieval system that finds the right Kaggle notebook snippets to use as few-shot prompts.