AI Leader · Scientist · Builder
Matthew Martz, PhD¶
I lead AI end to end: running research teams, shaping products, and building the engineering, data science, and platforms that put models into production. Over 22 years I've done this in cancer biology, consumer AI, biotech, agriculture, clinical medicine, and autonomy.
What I'm working on¶
-
Now: AI infrastructure at ASI
AI Infrastructure Engineering Manager at Autonomous Solutions, Inc. I lead hybrid compute, a company-wide lakehouse data platform, and the pipelines that take autonomy research to product.
-
Loom: healthcare AI platform
A precision-medicine platform built on knowledge graphs and patient digital twins. The knowledge graph covers 2.3M+ patients and 4.8B+ relationships, and federated learning lets sites collaborate without sharing data.
-
Research: ADAPT-Q
Specializing language models without making them forget: a family of fine-tuning methods that aims to learn a demanding new domain as well as standard approaches while preserving substantially more general capability.
-
Writing
Long-form technical posts on Python performance, data structures, MLOps in practice, and AI in biotech.
Research from Loom¶
Applied research on agentic AI, clinical knowledge discovery, and multi-domain learning, published on loom.clinic.
-
AI agents that discover and evolve what to optimize, not just how. Validated at Mayo Clinic for GraphRAG optimization, where it surfaced new levers such as citation density and temporal recency.
-
Agent-driven discovery of new conceptual units in clinical text (implicit relationships, meaningful absences, temporal sequences), guided by clinical outcomes rather than semantic similarity.
-
One model, many domains, without catastrophic forgetting: domain-specific adapters trained in orthogonal subspaces on a frozen base model, with router-based inference.
-
Production AI Systems at Scale
A compositional architecture for continuous learning: real-time adaptation without forgetting, domain-specific compliance and governance, audit-ready multi-domain systems, and 10× lower infrastructure cost.
-
Agent Evaluation Reference Guide
A structured reference for evaluating coding, conversational, research, RAG/GraphRAG, and computer-use agents: grader types, eval strategies, key benchmarks, and design principles.
Career at a glance¶
- 2026 – presentAutonomous Solutions, Inc.: AI Infrastructure Engineering Manager
- 2006 – presentLoom: Founder & Chief Architect (independent)
- 2026GiftHealth: Director of AI Engineering
- 2024 – 2025Mayo Clinic: Creator & lead, Agentic Knowledge Engine; Technical Lead & Solution Architect
- 2024Syngenta: Global Traits Digital Science Lead
- 2023 – 2024AgBiome: Head of Data Science / ML / AI
- 2017 – 2023Firstleaf: Head of Data Science / ML / AI
- 2014 – 2017UNC Lineberger Cancer Center: Postdoctoral Fellow, precision oncology AI
- 2006 – 2014Thomas Jefferson University: PhD, Biochemistry & Molecular Biology
Get in touch¶
I'm happy to talk about AI infrastructure, healthcare AI, knowledge graphs, or building and scaling AI teams. Email me at [email protected] or find me on LinkedIn.