Kartik
Journey
The Lab
Theme changes on reload
Ctrl+Kto search
~/kartik $ whoami

I build systems that survive contact with production.

I build AI agent infrastructure at Vertex, and I'm finishing my MS in Computer Science at Columbia. The problems I like are the ones where the answer has to actually run: on real data, on real hardware, with real users waiting.

New York, NYgraduating Dec 2026 · open to roles
See what I've builtPoke at the shellResume

01 / What I've been chasing

Most of my work starts as a question I couldn't stop poking at. Three that turned into something:

01

Can you trust what an agent just wrote?

TAOL scores every LLM-generated patch on four risk signals and routes it to auto-apply, human review, or defer. The ambiguity classifier hits 99.2% accuracy across 500 SWE-bench tasks. Turns out you can predict trustworthiness. You just have to measure the right thing, and an ablation study is what tells you which thing that is.

Read the build
02

What if privacy were structural instead of procedural?

UrbanistAI blurs faces and text before an image ever reaches storage, using three face detectors in a union ensemble because a false positive costs an over-blurred photo and a false negative costs someone's face. There is no code path that stores an unprocessed upload.

Read the build
03

How cheap can assistive hardware get?

Smart glasses for the visually impaired, 3D-printed parts, about $135 in materials. Sensor fusion for distance estimation and context-adaptive keyframe selection so the thing isn't burning compute on redundant frames. Two filed patents came out of it.

See the patents

02 / How I work

Measure before you claim

Instrumented every tool-discovery call at Vertex before touching a prompt. That's where the 36.11% token cut came from.

Tests are how you go fast

78 tests at Vertex, 144 on TAOL. A suite that catches regressions is what lets you keep changing things.

Hardware keeps you honest

The Pi and Arduino on my desk report live into this site. When they go down, you watch them go down.

Research should ship

Three papers and two patents, every one from an artifact I actually built and ran.

Currently
Vertex Inc.: MCP platform connecting AI agents to 105 enterprise tools across 10 providers, 36.11% leaner on tokens
Columbia aiX Lab: Evaluation metrics for data-science agents, plus AI literacy tooling

// off-hours: iced mochas, running, tennis, and whatever's half-built on my desk