Glenn Matlin
PhD Student • Georgia Tech • RAPID-AI
AI that thinks with people in socio-technical systems. I develop methods that let language models reason, analyze, and plan with humans for high-stakes decisions—grounded evaluations, training that improves reasoning, and assurance that earns trust.
RAPID-AI Research
T1: Grounded STS Evaluation
Building expert-level evaluations from authoritative sources—cited QAs and multi-step tasks that test knowledge, multi-hop reasoning, planning, and trade-off analysis across socio-technical domains.
T2: Training for Reasoning & Planning
Curated and synthetic data → SFT → RL with structure-aware rewards. Developing methods that improve generalizable reasoning rather than domain memorization.
T3: Assurance & Interpretability
Cognitive-pattern analysis, mechanistic and behavioral probes, uncertainty signals, and constraint checks. Making model behavior legible and auditable for high-stakes use.
T4: Human-in-the-Loop Interaction
Interfaces that elicit justifications, capture counterfactuals, and support red-teaming of plans. Ensuring humans can trust and appropriately use LM recommendations.
Featured Work
Research
FLAME - Financial LLM Evaluation
A comprehensive framework for evaluating large language models on financial domain knowledge, reasoning, and compliance tasks.
Latest Writing
The Challenges of Evaluating LLMs
An exploration of current challenges in LLM evaluation and promising approaches for more comprehensive assessment.
Publication
UnfoldML @ NeurIPS 2022
Cost-aware and uncertainty-based framework for dynamic 2D prediction in multi-stage classification systems.
Experience
Change Healthcare — Healthcare technology and analytics at scale
Komodo Health — Healthcare intelligence platform and real-world evidence
LendUp — Financial inclusion through responsible lending technology
RichRelevance — Personalization and recommendation systems at scale
:::
Let’s Connect
I’m always interested in discussing research collaborations, interesting problems, and new opportunities.