Hi, I'm Vaibhav

0 → 1

I was involved with Mapsense from before the company formally existed, joined full-time at seed as a founding engineer, and helped build the core product through its acquisition by Apple in 2015, about two years later. That set my defaults: small teams, direct ownership, product clarity.

Scale

Then roughly ten years at Apple Maps — ML systems, large-scale data infrastructure, geospatial systems, aerial imagery, production pipelines across 20+ countries. In my final year I designed the architecture for an internal agentic platform, and learned the lesson that now shapes my research: agents expose the debt a system already carries — fragmented data, knowledge living in people's heads, workflows that exist nowhere in machine-readable form. Real systems care about determinism, consistency, reproducibility, and reliability, and probabilistic components don't get a pass on any of it.

Back to 0 → 1

I left Apple in 2026 and moved to Indore. I'm working on the same question I kept running into at Apple, taken up without an organization in the way: how probabilistic AI becomes dependable in real systems, especially when it has to run local, offline, and small. I work on it as research and as advisory work with teams putting AI into real operations. The Acropolis Edge Lab — a student research cohort here — is where that question gets built and measured on actual devices.

Reading

Currently working through Richard McElreath's Statistical Rethinking. Favorites are on the homepage; more on Goodreads.

Working together

I'm looking for a small number of specific conversations: researchers and builders working on dependable AI, small models, edge systems, or geospatial and physical intelligence; founders and technical leaders who want a hard second opinion on an AI systems architecture; and universities or communities running applied research programs. Trained in operations research at Columbia, if you want the formal version: resume.

What I'm useful for →

Say hello or book a time.