GitHub

Vyoma Gajjar

I build production AI systems across machine learning, generative models, and agents.

My work spans the full loop: data and modeling, evaluation, deployment, monitoring, and the controls needed when probabilistic software meets the real world.

Current focus: tool-using agents, reliable multi-step workflows, and feedback loops that connect model behavior to system outcomes.

machine learning
NLP, predictive modeling, ranking, monitoring, drift detection, and governance.
generative AI
Retrieval, synthesis, evaluation, guardrails, and observability for systems built on foundation models.
agentic systems
Tool use, planning, routing, recovery, human handoffs, and the conditions under which a system should earn more autonomy.

agent-economics-lab

One current project focuses on the decision layer around agents. Once a system can complete a task, what evidence should determine whether it scales, gets assistance, or stops?

agent-economics-lab is a small, open-source, dependency-free decision engine that maps traces and outcomes to four actions: INCOMPLETE, SCALE, ASSIST, or STOP. The benchmark is intentionally mean to the evaluator: remove evidence and a naive cost optimizer confidently emits dozens of false SCALE decisions. The fail-safe policy emits zero.

A few things the internet can verify without trusting this page.

I occasionally speak about production machine learning, generative AI, agentic systems, evaluation, and responsible deployment.

Selected venues: PyData · Responsible AI Institute · IBM's Mixture of Experts podcast · IBM TechXchange (speaker and lab instructor)

For talks: gajjar.vyoma@gmail.com

Since 2021, I have taught machine learning, deep learning, production AI, RAG, generative AI evaluation, and responsible deployment through MIT Sloan (Great Learning), UCLA Extension, UT Dallas, and Maven.

I have spent the last decade building and operating AI systems, from NLP and predictive models to RAG pipelines and tool-using agents.

The through line is production: how to make models measurable, observable, steerable, and useful once they leave the notebook.

I am interested in systems that learn from feedback without hiding failure.

San Francisco Bay Area.