I am the co-founder of Headstarter, where I'm building autonomous coding agents that can ideate, build, deploy, and self-test web apps. Previously, I worked on ML at Amazon and data analytics at Bloomberg LP.
-
Hold Your Fire: Disruption-Aware Failure Monitoring for Coding Agents - Local, CPU-only failure monitor for coding agents with calibrated abstention, designed to detect failing trajectories early without interrupting runs that are still likely to recover.
-
Evidence Is Not Enough: Pass/Fail Signals That Don't Change a Coding Agent's Action - Mechanistic-interpretability study of pass/fail transcript evidence in static coding-agent prompts, showing how represented evidence can move internal margins without changing the selected action.
-
VANTAGE: Hidden Rewrite Views for Fixed-Prompt Speculative Code-Edit Decoding - Research prototype for fixed-prompt speculative code-edit decoding, using hidden rewrite views to recover copy-heavy draft spans while preserving target-model verification under the original prompt.
-
AgentRewind - TypeScript SDK for recording, replaying, inspecting, and forking LLM-agent runs across model calls, tool calls, and entropy boundaries.
-
metaharness - TypeScript-first SDK and CLI for running Claude, Cursor, Codex, and deterministic mock coding agents through one lifecycle, telemetry, policy, and handoff surface.
-
A Deep Learning Approach for COVID-19 & Viral Pneumonia Screening with X-Ray Images - Published research on computer-vision screening for COVID-19 and viral pneumonia from chest X-ray images.
-
My Journey to Big Tech: How I landed my internships at Amazon & Bloomberg
-
Death by a Thousand Handshakes: Why persistent connections are the future of AI agent infrastructure, and how one protocol change cuts agent latency in half.
-
True Alpha Lives in the Agent Harness: Everyone is optimizing the model. True alpha lives in the harness: the infrastructure that wraps the LLM and makes it actually work.