I build the AI systems — and the infrastructure they run on.

SCENE 01 / SYSTEMS IN PRODUCTION

• Webby Award winner
• First official Disney Theatrical podcast
• Ex-Google · drove the open-sourcing of Bazel · Developer Infrastructure at Google scale

calendarpa.com Live

CalendarPA

In a streaming org: Production scheduling at money-on-the-line correctness

Live · Public

Problem

Real scheduling spans Google, Microsoft, and Apple calendars, payments, and constraints ordinary tools can't express. Built solo — product, architecture, security, and launch.

Scale

50paying users in 45 days · zero double-bookings

Owner

Solo — product, architecture, security, and launch

Trust · how it holds up

AES-256-GCM encryption for sensitive availability at rest; distributed Redis locks on the booking critical section so money-on-the-line writes never double-book.

Tradeoff: Inngest durable jobs over a hand-rolled queue (observable background work in days, not weeks); a lock on the booking critical section over optimistic retries.

agentic AIencryption at restdistributed locksdurable jobsmulti-provider sync

See the deep dive Learn more →

calendarpa.com Live

AdaPA — agentic scheduling assistant

In a streaming org: A production AI agent taking real actions for real users

Live · Public

Problem

Most “AI assistants” summarize. AdaPA books: a LangGraph state machine with Gemini tool calling and pgvector RAG memory reads a plain-language request, checks real availability across every connected calendar, respects the rules the owner set — lunch, focus time, daily limits — and completes the booking over web chat, SMS, email, or MMS. It learns separately for each user from stated preferences, observed patterns, and booking outcomes, then adapts to how that person prefers to interact.

Scale

4channels: web, SMS, email, MMS

Owner

Solo

Trust · how it holds up

Every action runs through the same guarded booking path as the UI, so the agent inherits the distributed locks and encryption instead of bypassing them — it cannot double-book, because the invariant lives below it. Higher-risk channels add intent detection, forced identity lookup, tool-aware retries, and post-response integrity checks.

Tradeoff: The agent executes inside a constrained action surface rather than free-forming API calls — narrower than a general tool-use loop, but every path it can take is one the test suite already covers.

agentic AILangGraphpgvector RAGmulti-channelvision (MMS)

See the deep dive Learn more →

broadwaypodcastnetwork.com Live

Broadway Podcast Network

In a streaming org: Catalog ingestion, metadata freshness & offline downloads

Live · Public

Problem

A growing network needs a discovery platform that stays perfectly in sync with a third-party host, works offline on any device, and scales to peak — with no server to babysit.

Scale

200+podcasts in the catalog

Owner

Co-founder & CTO — led the design & system teams; sole maintainer today

Trust · how it holds up

AWS Cognito auth; a single guarded write-path keeps denormalized copies consistent, and relationship-heavy reads are served in one query.

Tradeoff: Megaphone as a headless CMS (one source of truth, no duplicate content system); a single-table DynamoDB model serves relationship-heavy reads in one query.

catalog & metadatafreshness SLAoffline / downloadsserverless scale-to-peakPWA

See the deep dive Learn more →

thebroadwaydropshop.com Live

The Broadway Drop Shop

In a streaming org: Premiere / on-sale surge — virtual queue & session integrity

Live · Public

Problem

High-demand timed drops melt naive checkout: everyone arrives in the same second, sessions break mid-purchase, and fairness matters — the same shape as a ticket on-sale, a premiere drop, or a live-event surge.

Scale

Orderedvirtual queue · capped concurrent checkout

Owner

Co-founder — solo technical designer and developer

Trust · how it holds up

A custom Lua admission script holds the queue order atomically inside Redis and caps how many buyers enter checkout at once; an HMAC-cookie + JWT dual session keeps a buyer's position through a mid-drop login.

Tradeoff: A custom Redis virtual-queue over a hosted waiting room — drop fairness and timed admission needed precise control of position and reaping that hosted rooms don't expose.

live-event surgevirtual queuesession integrityfair admissionLua

See the deep dive Learn more →

levelmyaudio.com Live

LevelMyAudio

In a streaming org: Media processing — loudness compliance on elastic workers

Live · Public

Problem

Professional audio leveling is slow, expensive, and gatekept by expertise. Creators need broadcast-quality output (−16 LUFS) without an engineer in the loop — the audio analog of an encode/QC pipeline.

Scale

10×reduction in the manual workflow

Owner

Solo

Trust · how it holds up

An SQS + Fargate async pipeline survives long encodes without blocking the UI, and every job is measured against the −16 LUFS broadcast target rather than judged by ear.

Tradeoff: Fargate over always-on workers (bursty jobs, no paying for idle); an async job + notification over synchronous processing (long encodes can't block a request).

media processingloudness / QC−16 LUFSelastic workersTerraform

See the deep dive Learn more →

bazel.build Live

Bazel — Google build platform

In a streaming org: Developer productivity / build platform — at Google scale

Live · Public

Problem

An estimated 50,000 internal Google engineers needed to build at Google scale without waiting on each other. As the Technical Program Manager on the Developer Infrastructure team, I drove the open-sourcing of the build system that solved it — taking Google’s internal build platform out into the world as Bazel.

Scale

50,000internal Google engineers at the time

Owner

Technical Program Manager — drove the open-sourcing

Trust · how it holds up

The launch outlived the role: Bazel went on to be widely adopted worldwide, extended by a community-written rules ecosystem, and maintained in the open rather than as an internal fork.

Today

A decade on, Bazel is a flagship open-source build system — “{ Fast, Correct } — Choose two” — building Java, C++, Go, Android and iOS across Windows, macOS and Linux, extended by a community-written rules ecosystem, and stewarded by its own Build Foundation.

Tradeoff: Open-sourcing internal infrastructure over keeping it internal: far more work to make a Google-shaped build system usable outside Google, but it is why the thing still exists a decade later.

developer productivitybuild platformremote executionmonorepoopen source

Visit bazel.build

Deep: distributed systems · serverless at peak · build / developer productivity · media (audio) pipelines · audio & video editing · AI / RAG / agents. Extensible: video transcode & packaging · CDN / QoE.

SCENE 02 / TELEMETRY + TRUST

• Encryption at rest

  AES-256-GCM on sensitive availability data — a default from day one, not a later retrofit.

  CalendarPA

• Audit logging

  Every financial mutation is logged, behind a QuickBooks OAuth 2.0 integration.

  P&L automation

• Concurrency safety

  Redis distributed locks on the booking critical section — money-on-the-line writes never double-book under load.

  CalendarPA

• Fair admission under load

  A custom Lua admission script holds queue order atomically inside Redis and caps concurrent checkout, with an HMAC-cookie ↔ JWT dual session that survives a mid-drop login so a buyer’s hard-won position holds.

  The Broadway Drop Shop

• Failure isolation

  Async SQS + Fargate pipelines survive long jobs without blocking the UI; a single guarded write path keeps denormalized state consistent.

  LevelMyAudio · BPN

• Test rigor

  Over 10,000 unit tests completing in under 30 seconds, with continuous integration and gated production security mitigations.

  CalendarPA

SCENE 03 / THE LOG

1. CalendarPA

   2026 — Present

   • Founder & CEO

     2026 — Present

2. Broadway Podcast Network

   2024 — Present

   • Co-Founder & CTO

     2024 — Present

3. Google

   2013 — 2024

   • TPM, Developer Operations & Infrastructure

     2013 — 2016

   • Program Manager & Global Lead, Talks at Google

     2016 — 2022

   • Program Manager, GRAD

     2022 — 2024

4. Thomson Reuters

   2010 — 2013

   • Senior Network Engineer

     2010 — 2013

5. USA Studios

   2008 — 2010

   • Director of Information Technology

     2008 — 2010

SCENE 04 / OPEN CHANNEL