Ajay nagesh
NLP/AI Researcher — LLM Evaluation · Agentic Systems · Representation Learning
[publications] [resume] [LinkedIn] [writing]
Embedded Files
I am a Senior Applied Scientist at Amazon (Sunnyvale, CA), working on LLM evaluation and agentic systems.
I was previously Staff Research Scientist at DiDi AI Labs and postdoctoral researcher at the University of Arizona and UMass Amherst. I hold a joint PhD from IIT Bombay and Monash University.
My research spans LLM evaluation methodology, agentic systems, semi-supervised learning, representation learning, and information extraction.
Research Writing
Research Writing
On the hard problem of evaluating AI — LLM-as-judge methodology, analysis-first prompting, and what systematic evaluation requires.
Retrospective on agentic AI through the lens of a 2020 ride-hailing agent — what the field has caught up on, and what it hasn't.
Practitioner's notes on the data problem — filtering noisy corpora, augmenting limited labels, and building structured synthetic pipelines.
On semantic drift, Ladder Networks, and why the EMA stabilization trick at the heart of Mean Teacher keeps reappearing in modern ML.
On emergent geometry, interpretable representations, and what 2018 NEC research says about modern SAEs and the linear representation hypothesis.
Explorations
Explorations
Working through Karpathy's nanochat repository — annotated observations on modern LLM architecture.
A workflow for understanding the attention mechanism — visual, hands-on, and practical.
Page updated
Google Sites
Report abuse