I started in enterprise firmware at Micron, where I held a sub-1-second SLA on activation for 4TB to 128TB SSDs shipped to Fortune 500 clients, then moved to Amazon and cut device downtime 30% across 150+ fulfillment centers. At Micron I also recovered approximately two weeks of recurring manual effort and reduced onboarding time 75%, mentoring 10 engineers along the way.
Now at NYU Stern's Endless Frontier Labs I own the US market-entry workstream for an AI platform, building bottom-up TAM/SAM/SOM models and translating architecture gaps into prioritized roadmaps for founders. In parallel I have open-source tooling for LLM agents—capability-based permission engines, OAuth gateways for the Model Context Protocol, and content-addressed pipeline runners—and data platforms that process 50M+ rows a month.
| Initiative | Scope | Outcome |
|---|---|---|
| US Market Entry for AI Securitization Platform | Bottom-up TAM/SAM/SOM model, AI architecture evaluation, data readiness assessment | Prioritized roadmap for founder and product team with target segment recommendation |
| Enterprise SSD Firmware Activation | Sub-1-second SLA window for 4TB–128TB drives at Dell and HP | Zero-disruption upgrades in live production environments |
| Fulfillment Center Device Reliability | Root-cause diagnostics across 150+ centers, new error-to-success ratio metric | 30% reduction in device downtime |
| Test Automation Framework | Centralized framework standardized across 100+ concurrent setups | Recovered approximately two weeks of recurring manual engineering effort |
| Engineer Onboarding Program | Knowledge base and curriculum for 10 engineers | 75% reduction in time-to-productivity |
| agentguard | Capability-based permission engine for LLM agent tool calls with YAML policies, argument constraints, taint tracking, human approvals, and hash-chained audit log | Open-source security infrastructure for agentic workflows |
Behind the Fulfillment Center Device Reliability initiative
I drove a 30% reduction in device downtime across 150+ fulfillment centers.
- MS Engineering Management, New York University, Sep 2025 – May 2027 · GPA 3.80 / 4.00 · GMAT Focus 655
- Integrated B.Tech + M.Tech ECE, IIIT Bangalore, Aug 2018 – Jul 2023 · GPA 3.50 / 4.00 · Top 1% nationally in JEE
- agentguard — Capability-based permission engine for LLM agent tool calls, with YAML policies, argument constraints, taint tracking, human approvals, and a hash-chained…
- mcpgate — A Python security gateway for the Model Context Protocol that fronts many MCP servers with OAuth bearer validation, per-client tool RBAC, rate limits, audit…
- labelbudget — An active-learning benchmark that reports labels needed to reach a target accuracy with variance across many seeds, beside a prominent random-sampling baseline.
- fatigue — Models ad creative fatigue as a hazard function of cumulative per-user frequency rather than calendar time, and predicts a refresh date per creative with a…
- cannibal — Search cannibalisation detection that tests page pairs with multiple-testing correction, so the output is a short list of real competing pairs instead of…
- dagcache — A content-addressed, parallel pipeline runner for reproducible ML workflows that hashes code, params and data, explains staleness and traces artifact lineage.
- nyc-taxi-data-lake-gcp — Processes the NYC TLC Trip Record Dataset (50M+ rows/month) with Dataproc + PySpark, writes optimized Parquet to GCS with Hive-style partitioning, serves via…
- weather-data-pipeline-great-expectations — A production-grade weather data pipeline that collects hourly observations from 50+ global cities via the Open-Meteo API, validates every batch with Great…
- planprice — An offline Go CLI that prices Terraform plans and diffs cost between plans from a pinned AWS price snapshot, with PR-ready Markdown and CI cost gates.
- spotify-listening-analytics — A full-stack data engineering platform that extracts personal Spotify listening history, playlist data, and audio features via OAuth 2.0, builds a dimensional…

