I'm a staff AI product manager at Arize, working on the loop that makes agents better over time: observability, evals, and experimentation. I lead product strategy and our self-serve motion, and I still ship PRs when I can get away with it.
Before product I was an engineer. Flight software at SpaceX (Mars program, Dragon), the mobile app and Virtual Power Plant at Tesla, content moderation systems at Twitter. Then Federato, where I moved from senior engineer to co-leading product.
Outside work I co-host The Next Iteration with Damien Joseph, and I've started publishing research on how institutions turn what people want into decisions, in parliaments and in multi-agent AI systems.
I grew up crossing borders (Austin, Laredo, Bangladesh, Toronto, Texas again, the Bay Area, now Brooklyn), which is roughly why this site is a map.
Built and ran the post-training loop for Alyx, our in-product agent: evals, verifiable rewards, SFT, and RL (GRPO), reducing inference cost by 30x and latency by 9x while increasing accuracy from 64% to 83%. Launched agent experimentation, shipped Arize's first REST API, MCP, and CLI releases, and grew API traffic from 10K to 4M requests per week. Led the Managed Agents team. Joined as a PM, promoted to Senior PM in February 2026, then Staff PM in September 2026. Arize was acquired by Dynatrace for approximately $1 billion in August 2026.
Came in as a senior engineer and moved over to co-lead the product org. Led the homepage redesign end to end through user interviews, PRDs, and engineering execution, growing DAU 1048% from launch. Spearheaded pricing and packaging strategy that enabled a 67% ACV increase. On the engineering side, launched a production instance for a 100+ user climate insurance client in eight weeks, leading seven engineers across the risk platform and Kafka, Grafana, Python, and React data pipelines.
Sourced and evaluated early-stage startups with the Susa team, early investors in Robinhood and Flexport. Angel invested in deep tech through the Susa Fellows Fund.
Engineered a content moderation system for Tweets, improving the efficiency of hate-speech and harmful-content removal. Built large-scale distributed systems in Scala to batch review reported media files, 100TB+ of data, at global scale.
Coded and installed a network of weather-station monitors to create a prediction system for optimal launch times. Built classified, mission-critical flight software for the Mars program, Starship, and Dragon, including path planning for robotics integration and real-time telemetry.
Cut Tesla mobile app load times by 20% with clustered Scala and Kubernetes microservices. Worked on reliability for the Virtual Power Plant and Autobidder.
Institutions decide whose preferences survive the process. I study that in two places, democratic legislatures and multi-agent AI systems, using simulations where the right answer is known.
A reviewer can repair a bad proposal or destroy a good one. Across two open-weight models and two task domains, restricting override authority sharply reduced corruption, but sometimes blocked useful improvements. An explicit retain option eliminated corruption for Qwen but not for Mistral.
Fragmented parliaments don't collapse because coalitions fail to form. With MPs voting personally, passage (46.4%) matches the presidential benchmark (44.8%); collapse to 0.05% requires a cohesive opposition.
The Next Iteration
"Enabling you to build the best iteration of yourself"
Damien Joseph and I interview founders, operators, and investors about how they built what they built, and what they'd do differently.
A Jev side experiment: following people's priorities, changing option order, and checking the probabilities.
Why reviewer quality is only half the problem when one AI agent can overrule another.
Three surprises from 3,200 simulated legislative sessions across four democratic institutions.
On maps, borders, and why I turned my portfolio into terrain you can explore.
Email is the fastest way to reach me. I'm up for talking about agent evals and observability, podcast guests (pitch me someone), or anything on this map.
brooklyn, ny. usually online.