Tim Bydlon

Advanced planning system for agentic tool use, extending my research at DILab. Implementing sophisticated algorithms for AI agents to efficiently plan and execute complex multi-step tasks.

PyTorch implementation for estimating incremental treatment impact on customer outcomes using causal inference. Predicts which customers will respond best to marketing campaigns using GRU-based sequence modeling and doubly-robust learning.

Docker-based fine-tuning environment for large language models using LLaMA-Factory on NVIDIA DGX. Implements efficient LoRA adaptation techniques for Llama-3 models with comprehensive experiment tracking and batch inference capabilities.

Sample implementation of semantic search using Pinecone. Demonstrates experience with embedding-based retrieval systems used in both DILab research and Salesforce production systems.

Model Context Protocol server enabling secure local LLM data analysis. Demonstrates a sample of my work building MCP integrations for enabling conversational queries and improving workflow efficiency.