Senior Software Engineer at AWS — Search Services · Data Prepper, OpenSearch Ingestion.
I build systems that have to be fast, and products that have to be honest about why they say what they say.

A free, interactive book about data structures as the machine actually runs them.

Two loops, same operation count, same O(n). One of them is thirty to sixty times slower — because of the order in which it touches memory. Most data structures books stop at Big-O. This one keeps going: cache lines, prefetchers, memory layout, false sharing, B-trees, LSM-trees, concurrency. That's where the performance of real systems is actually decided.

→ data-structures-on-systems.vercel.app

• Live benchmarks that run on your CPU. The memory hierarchy chapter measures your machine while you read it — latency at every cache level, the real cost of a miss, what array-of-structures charges you. No screenshots of someone else's laptop.
• "Watch it run" visualizations. Step through two-sum, Kadane's, a BST search, a B-tree split, an LSM flush/compaction, a thread race vs. a mutex, a BFS traversal — animated, at your pace.
• Runnable problems in four languages. C++, Python, Java, Go — compiled and judged in the browser, including the practice set.

Astro · MDX · WebAssembly · MIT licensed → sb2k16/data-structures-book

The companion app to the book. 70 algorithms implemented in C++ — not simulated in JavaScript. The engine runs the real code and emits one NDJSON event per step; the browser replays that trace, so the animation you scrub through is the execution, line-synced with the source.

Around the player: an AI tutor grounded in the book (RAG over the chapter text), a mock-interview mode, a weekly themed problem marathon with a leaderboard, and a mastery model that tracks concepts rather than problems solved and diagnoses what's actually weak. Code runs in six languages against real compilers.

C++ · Express · React · Vite · Expo · Clerk · Neon · Anthropic · Vercel

Automates the grad-school outreach pipeline for prospective Mathematics and Computer Science applicants. It ingests a student's manuscript, extracts its mathematical footprint into a semantic research vector, matches that vector against thousands of active faculty, flags reviewer-risk gaps in the work, and drafts precise cold outreach.

Behind it: an OpenAlex-backed corpus of ~2k faculty and ~70k papers per discipline, embedded and searched with pgvector centroids, plus an NSF grant pipeline that scores each lab's funding health.

Express · TypeScript · Neon Postgres + pgvector · Clerk · Anthropic · Fly.io · Vercel

Learns how an engineer thinks across three pillars — Algorithms, System Design, AI Engineering — and continuously answers one question: what is the single best thing I should learn next?

Built around a knowledge graph and a three-tier skill detector, so it shows the why behind a score, not just the score. Design closer to Linear/Cursor than to Coursera.

React · Express · Neon Postgres · Clerk · Anthropic

A community-first math app: curated problems worth thinking about, rather than an endless drill feed. Ships with procedural generators (~7,700 problems) behind a hand-built design system, on web and native.

Next.js 16 · React 19 · Expo · KaTeX · Clerk · Neon

One-tap live drive times, target-arrival alarms that back-calculate when to leave (departBy = arrival − liveDuration − buffer) on a 10-minute cron, and an SMS bot that answers status with your current commute. Multi-stop optimizer deep-links straight into Google Maps.

Next.js · Expo · Google Routes API · Twilio · Vercel Cron

Presence at the vulnerable moment, not streak-counting. Starts with nicotine/vaping.

The architecture is a product argument: safety is the spine, not the ER. Every inbound message is screened for crisis language before anything else runs, and every word the model produces passes an output guard that can never assert a reassuring risk verdict as fact. All intelligence is server-side, so the guard cannot be bypassed by a client.

Express · Expo · Clerk · Neon · Anthropic

A 24/7 conversational companion, wellness and companionship rather than treatment, with the same non-bypassable safety spine intercepting crisis on every turn and routing to real help. Runs end-to-end keyless in mock mode — no API keys, no database — so the full loop is testable by anyone.

Express · Expo · TypeScript

Preference-aware meal planning with a shopping cart and calorie calculator on a Go + Node backend.

Go · Node · React · Docker

Scenario-driven, YAML-defined evaluation for LLM apps, as a CLI (aitest) and a dashboard. Because "it looked right when I tried it" is not a test suite.

TypeScript · Anthropic SDK · Zod · Express

_{Most of the product repos above are private while they're in flight — the live links are the real thing, not screenshots.}