Shopper Agent Evaluation
I designed and led the evaluation framework for Shopper Agent. Simulated conversations and automated checks help the team test changes across storefronts and ship with more confidence.
Applied AI and machine learning
Since 2019, I’ve worked across the full ML lifecycle, from developing predictive models and AI agents to deploying, evaluating, and improving them in production. Much of that work has been in enterprise SaaS, including revenue intelligence at Clari and agentic commerce at Salesforce.
I’m interested in how agents are designed, what context they need, and how we can tell whether they’re getting better. That includes testing realistic scenarios in simulation, checking that evaluations are reliable, and learning from what happens after launch. I also care about what makes AI features and models useful enough for customers to adopt and reliable enough to keep using, especially when a shared platform has to support different data and workflows at scale.
This is where I share projects I’ve worked on, ideas I’m thinking through, and notes on what I’m learning.
I designed and led the evaluation framework for Shopper Agent. Simulated conversations and automated checks help the team test changes across storefronts and ship with more confidence.
I led the work to add conversation and meeting signals to Clari’s deal-scoring model. The enhanced score helped sales teams spot at-risk deals and decide which opportunities to prioritize in pipeline reviews.
I built a model registry that supported the lifecycle of production AI and ML features for more than 1,000 customers.
How we evaluate a shopping agent across storefronts and test the evaluators themselves. With Pushkar Garg.
Notes on model serving, multimodal data processing, and what I wanted to hear about agent evaluation.
Using dependency parsing to extract phrases from customer reviews, even when the words are apart.