Staff engineer · AI agents & full-stack product

Michael Nakayama

Software engineer with 11 years across startups and scale-ups, currently building production LLM agents at MagicSchool AI. I lead cross-team projects from prototype to 100% rollout, and I care about measuring outcomes with evals, experiments, and analytics.

Selected work

Now

Experience

11 years

View

Focus

Jan 2025 – PresentRemote

MagicSchool AI

Staff Engineer

2026

Raina AI teaching assistant

Rebuilt MagicSchool's AI teaching assistant as a tool-calling agent and rolled it out to all production traffic.

conversations / week
1.1M
of traffic, zero rollbacks
100%
tools callable from chat
100+

2025–2026

Studio Mode

Designed and built, on my own, the collaborative document workspace inside Raina where teachers generate and edit lesson plans, worksheets, and other classroom documents.

documents / school day
70K+
assistant turns create or edit a doc
1 in 4

2025

Data Analysis Tool & platform

Took a conversational data-analysis product from zero to launch and prototyped the agent infrastructure the 2026 work built on.

Feb 2024 – Nov 2024Remote

Pursuit

Founding Full Stack Engineer

One of the first engineers; built an ML forecasting product that signed three customers in its first month.

revenue in month one
$100K+
of records synced to CRMs
Millions

Mar 2022 – Feb 2024Remote

Olo

Senior II Engineer

Scaled the event platform behind restaurant ordering and built a customer reporting engine on top of it.

events / day
1M+
throughput increase
5x
data points kept accurate daily
300K+

Aug 2015 – Mar 2022Bellingham, WA

Faithlife

Senior Engineer · Team Lead

Led a church-management platform from design to production and managed the team building it.

churches on the platform
100+
direct reports
4–8

Projects

3 in 2026

2026

Fantasy football decision engine

Python · LightGBM · PyTorch · scikit-learn · FastAPI · Next.js · Modal

A personal ML system for fantasy football: a season simulator, draft engine, and trade model behind a FastAPI backend, Next.js dashboard, and MCP server, deployed on Modal. I write about the models on mnaks.me.

rows of simulated seasons
600M+
rows of NFL data ingested
4.2M
real drafts crawled
6.6K

2026

gemmr: trading-card grading

Python · OpenCV · PyTorch · Modal · Supabase

Computer vision for a collaborator's trading-card grading product, built into an existing production codebase.

2026

MLB Money

Python · SQLite

No-run-first-inning and matchup analytics built on the free MLB Stats API.

Skills & education

  • Skills

    Languages
    TypeScript, JavaScript, Python, SQL, Go
    AI
    LLM agents & tool calling, Vercel AI SDK, MCP, Amazon Bedrock, Braintrust evals, Prompt engineering
    ML
    PyTorch, LightGBM, scikit-learn, OpenCV, Monte Carlo simulation, Probabilistic forecasting & calibration
    Frameworks
    React, Next.js, NestJS, Node.js, FastAPI, Playwright
    Data & infra
    Postgres/Supabase, AWS (ECS, Lambda), Trigger.dev, Temporal, Elasticsearch, Redis, Datadog, Amplitude
  • Education

    B.S. Computer Science
    Vanderbilt University, 2015