AI and Machine Learning Developer

Mar 2026 — Present

• Build the conversational booking agent: a planner that fans out parallel specialist workers and streams generative UI into chat, with compact worker summaries that cut model context by about 60%.
• Designed the self-healing tool contract: supplier failures become retry hints and visible diagnostics instead of silent errors or raw stack traces.
• Made chat streams durable on Cloudflare Workers, so generation survives disconnects and resumes across isolates during 20-second multi-supplier searches.
• Audited and fixed around 50 vulnerabilities across auth, payments, race conditions, and multi-tenant isolation.
• Own the eval harness (deterministic assertions plus cross-family LLM judges) and ship the MCP server that lets assistants like Claude search and book hotels.

Sep 2025 — Dec 2025

• Ran hands-on workshops on AI/ML, the web platform, and cloud for the local developer community.
• Mentored students and early-career developers through their first open source contributions.

Jan 2022 — Sep 2024

• Ran technical direction for AI/ML, platform, and infrastructure, and grew the engineering team from zero.
• Shipped production ML pipelines and web products that make technology easier for people to use.
• Designed the architecture for data ingestion, model serving, and the customer-facing web stack.

2021 — 2024

• Foundations in algorithms, systems, databases, networking, and software engineering.
• Graduated with a focus on applied machine learning and full-stack web development.

2024 — 2025

• Graduate coursework in deep learning, distributed systems, and advanced algorithms.
• Research interests: neural networks, NLP, and scalable ML systems.

2025 — Present

• Self-hosted personal mail server in Go: IMAP, SMTP, CalDAV, CardDAV. A privacy-focused drop-in for Gmail/Outlook.
• Production hardening: auto-TLS, DKIM/SPF/DMARC signing, greylisting, Prometheus metrics, structured audit logging.

2026 — Present

• macOS menu-bar app that lets you switch the default browser from the system tray instead of digging through Settings.
• Native Swift, sub-second open-to-switch flow, zero background CPU when idle.

2025 — Present

• Encrypted, versioned `.env` storage per project and environment: upload, diff, and roll back without leaking secrets to git.
• Next.js + tRPC + Drizzle on Turso, with client-side encryption so the server never sees plaintext.

If one book explains how I operate, it is this one. Ship early, measure, learn, repeat. Most of what I shipped too early, I shipped because of this book. No regrets.

The closest thing to a manual for my day job. Serving models in production is its own discipline and this is one of the few books that treats it that way.

I became a CTO while still in undergrad. This is the only startup book that felt like it was written by someone who had actually been in the room when things went wrong.

Carr was right about what the internet does to attention. I write long articles partly to prove to myself I can still think in long form.

2017

The obvious pick, but I reread it every year or so and still catch details I missed. Everything below sits on top of this.

2022

The transformer paper tells you what attention computes. This one tells you why it was slow, and fixes it by caring about memory movement instead of FLOPs. It changed how I look at GPU cost.

2023

The sequel. Reads like a changelog of lessons from running v1 at scale: better work partitioning, fewer non-matmul ops, and a big jump in occupancy.

2024

The point where kernel work becomes hardware co-design: async warp specialization and FP8 on Hopper. You stop optimizing for a GPU and start optimizing with it.

2023

The vLLM paper. Paging the KV cache like an OS pages memory is the kind of idea that looks obvious only after someone does it. Made serving feel like systems engineering instead of magic.

2023

RadixAttention makes prefix caching a first-class data structure instead of a lucky cache hit. The radix tree over KV cache is the trick I wish I had thought of.

2022

Draft with a small model, verify with the big one, keep the tokens they agree on. Free speedup with the exact same output distribution. Still feels like cheating.

2024

Speculation without a separate draft model, just extra decoding heads on the model you already have. Practical and a little audacious.

2024

The strongest take on speculation I have read: predict at the feature level, not the token level. The draft model gets the hard part for free.

2023

Prefill and decode want different hardware, so run them on different machines. Obvious in hindsight, which is the best kind of idea.

2024

The scheduling math behind prefill/decode disaggregation, with goodput under latency SLOs as the metric that actually matters. Closest paper to my day-to-day latency budgets.

2023

Quantization that respects which weights actually matter by looking at the activations, not the weights themselves. Protect the salient 1% and the rest can be cheap.

2022

The paper that made 4-bit weights respectable. One-shot, no retraining, and the error compensation is proper old-school numerical thinking.

2022

Activations are hard to quantize and weights are easy, so migrate the difficulty from one to the other ahead of time. Elegant.

2026

The newest addition to the stack. Sub-1-bit sounds absurd until you read how they share structure across weights. I am still deciding what I think.

2023

I run LLM judges in production evals and this paper is honest about their biases. I learned the same lesson the hard way: do not let a model grade its own family.