Joel Brelson

I’m a Machine Learning Engineer and AI Research Engineer building Generative AI and LLM-driven systems for drug discovery and bioinformatics. At Boltzmann Labs, I improved de novo molecular design models, cutting inference time by 30% through optimized architectures and sampling, and I delivered retrosynthesis pipelines with 150% more valid pathways and 25% shorter routes.

I designed SynAgent, a modular retrosynthesis system (11 tools + LLMs) and an uncertainty-aware feasibility model using RXNBERT fingerprints with 480k+ training examples, integrating calibrated uncertainty into route ranking and deploying cost-efficient 4-bit quantized LLMs. I also built multi-agent and Agentic RAG architectures (110+ tools) using LangChain/LangGraph, reducing input tokens by 60% while improving multi-hop reasoning accuracy, and I containerized ML workflows with Docker & AWS EKS to scale the REBOLT platform to 100+ concurrent users. Previously, as an intern, I developed a U-Net–based protein binding-site predictor, improving accuracy by 20%.