principal engineer · 20 years in production · denver, co
➜ ~ cat bio.txt
I’m a hands-on engineer who keeps high-scale systems alive. Right now I architect AWS-native infrastructure handling hundreds of millions of daily requests for a real-time gaming platform — infrastructure as code, event-driven services, and data pipelines. After years leading as a CTO, I’m back where I do my best work: on the hardest problems.
The engineers who pull ahead won’t be the ones AI replaces — they’ll be the ones who wield it well. I treat models as production infrastructure: gated, non-blocking, and human-in-the-loop — never a bolt-on, never unsupervised. Used right, that leverage lets one engineer ship what used to take a team.
01
Leverage over headcount
AI-augmented workflows let a small, senior team — or a single engineer — ship what used to take a floor of engineers. I design the pipelines, prompts, and guardrails that turn that leverage into reliable throughput.
02
Judgment stays human
Models handle the toil — boilerplate, migrations, first-draft code and docs. Architecture, tradeoffs, and the call on what ‘good’ means stay where they belong: with the engineer.
03
LLMs as production systems
An AI feature is a distributed system like any other: async pipelines that never block the app, versioned prompts and models, observability, cost and latency budgets, and failure modes planned for up front.
04
Raising the team’s floor
The real multiplier isn’t my own output — it’s the tooling, standards, and review practices that make every engineer around me faster and safer with AI.
// in production — how I actually ship AI
Leverage, proven
A fleet of production and near-production Claude systems — a live trading optimizer, an agentic inbox brief, an AI video pipeline, a fitness coach — built and run largely solo. That breadth is only possible with AI leverage.
Gated, never unsupervised
Every model runs behind a gate. Claude annotates pull requests but never blocks a merge; a trading optimizer escalates suggest → paper → live only past a confidence bar, behind a circuit breaker; coaching stays descriptive, never prescriptive.
AI that never blocks
LLM calls are async, budgeted, and fallback-safe — a recording plays whether or not its AI summary is ready, and a two-tier Haiku/Sonnet split trades cost against depth per call.
> AI-first isn’t a tool you adopt. It’s how you architect, review, and ship — end to end.
01
Selected Work
// case studies
Consulting & roles// track record
Products & AI// shipped — most live
Open source & tooling// public
Hijack PokerCTO / Principal Architect
Real-time gaming platform at 300M requests/day
Architected and operate AWS-native infrastructure for an enterprise online poker platform — spanning online and brick-and-mortar operations, running 24/7. Built the real-time layer, the IaC, and the data engineering behind it, and lead an 18-person technology org.
CI/CDGitHub Actions→Automated Tests→Blue / Green Deploy→Backend + Unity clientsprovisioned via Pulumi
Kalshi-BotPersonal · AI-optimized trading system
A self-tuning fleet of prediction-market trading bots
A production fleet of algorithmic bots trading Kalshi’s short-window crypto prediction markets on momentum and trend signals — with Claude wired in to read live performance and propose strategy refinements. Containerized, deployed to AWS, running 24/7.
A daily agent that reads Email and Slack from the last 24h and synthesizes a ranked brief — urgent, committed, FYI — with Claude Sonnet. Each item opens a chat thread where the agent drafts tickets and replies.
Claude SonnetAgent ChatLaravelVue 3PWA PushAWS
// BRIEF PIPELINE — INGEST → SYNTHESIZE → ACT
SOURCES
Email · Slack
last 24h
→
SYNTHESIZE
Claude Sonnet
rank · cluster
→
BRIEF
3-Tier Digest
urgent · committed · fyi
⇢
ACT
Agent Chat
draft tickets / replies
DELIVERYQueued synthesis job→6am push→PWAan agent per item
jira-mcp
Open tooling · Model Context Protocol server
JIRA project intelligence, exposed as agent tools
An MCP server that exposes JIRA project health as tools any LLM client can call — surfacing blocked, stalled, and ‘bouncing’ tickets, at-risk epics, and transition bottlenecks for agentic project monitoring.
An open-source stack that puts whatever you’re building on a real HTTPS URL straight from your own machine — through a single outbound Cloudflare Tunnel, with Caddy routing each hostname to the right container. No server, no open ports, no deploy.
A fitness app with a conversational AI coach — Claude Sonnet with multi-turn memory and coaching personas — plus body-composition analysis from a photo (Claude Vision, through an S3 → SQS → Lambda pipeline). A real-time, on-device form coach is in progress.
Claude (Sonnet + Vision)LaravelVue 3CapacitorAWS LambdaSQSMediaConvert
// AI PIPELINE — CAPTURE → PROCESS → ANALYZE → COACH
CAPTURE
Photo / Video
mobile · Capacitor
→
PROCESS
S3 · SQS · Lambda
async · MediaConvert
→
ANALYZE
Claude Vision
body-fat % · quality
→
COACH
Claude Sonnet
memory · personas
⇢
BUILDING
Real-time form coach
MediaPipe · on-device
SHIPPEDPhoto body-comp→Conversational coach→HLS videoreal-time form coach in progress
A multi-tenant screen-recording platform — browser capture uploaded direct to S3 (the app server never touches video bytes), MediaConvert transcode to HLS, and a Claude + Transcribe pipeline for transcripts, summaries, chapters, and action items. In use internally and by external customers.
A cross-platform tracker for people running peptide, GLP-1, and hormone protocols — it remembers your regimen, schedules reminders, and handles the dosing and reconstitution math. It logs injections and bloodwork and imports Apple Health / Health Connect data, shipped as web plus native iOS and Android from one codebase.
LaravelVue 3CapacitorStripeApple Health / Health ConnectPostgreSQLRedis
// ARCHITECTURE — CLIENTS → API → ENGINE → DATA
CLIENTS
Web · iOS · Android
one Capacitor build
→
API
Laravel
api.pepme.ai
→
ENGINE
Protocol Engine
dosing math · schedule
→
DATA
Postgres · Redis
protocols · reminders
DELIVERYShared Vue + Capacitor→Push reminders→App Store + Playremembers · reminds · does the math
Maestro
Content platform · AI-assisted
Turning screen recordings into living documentation
A content-transformation platform that migrates, creates, and enhances documentation across SharePoint, Dozuki, and Confluence — including a desktop recorder that captures a workflow and turns it into a step-by-step SOP, with a versioned document-lifecycle engine underneath.
Architected the medallion data lake & warehouse strategy for a utility serving 3.7M+ consumers — Apache Iceberg tables, a two-phase CDC pipeline (row-hash updates, table-diff deletes), and a zero-downtime cutover from legacy Oracle.
Delivered security remediation across production apps and AWS environments — zero-trust microsegmentation with Illumio, Cognito/IAM identity hardening, encryption at rest and in transit, and an automated security-test suite.
IAM / VPCCognitoIllumio (Zero-Trust)EncryptionTest AutomationTerraform
// DEFENSE-IN-DEPTH — IDENTITY → NETWORK → DATA → AUDIT
IDENTITY
Cognito
OAuth2 · MFA
→
EDGE
API Gateway
security policies
→
NETWORK
Illumio
microsegmentation
→
DATA
Encryption
at rest / in transit
⇢
AUDIT
CloudWatch
logging · detection
SECURE CI/CDDependency scanning→Container hardening→Automated vuln scansleast-privilege IAM
Dun & BradstreetLead Solutions Engineer
Enterprise API platform for commercial data
Led a team of 6 building a microservice API ecosystem for business-intelligence products — OAuth2 auth, Redis caching, rate limiting, and a versioning strategy for public endpoints.
100K+
requests/day
6
engineers led
MicroservicesOAuth2RedisAPI VersioningBatch Sync
// API ECOSYSTEM — CONSUMERS → GATEWAY → SERVICES → DATA
CONSUMERS
API Clients
100K+ req/day
→
GATEWAY
API Gateway
OAuth2 · rate limit
→
SERVICES
Microservices
BI data products
→
CACHE
Redis
high-freq reads
→
DATA
Data Stores
batch sync
PLATFORMAPI versioning→Deprecation workflow→Abuse detectionteam of 6 engineers
Prologis
Senior Software Engineer
Event-driven platform for supply-chain ops
Built an event-driven microservices platform for warehouse and supply-chain operations — decoupled services over a message backbone, with fault-tolerant integrations to third-party logistics systems.
RELIABILITYIdempotent handlers→Dead-letter queues→Automated retriesfault-tolerant by design
Whole Foods Market
Senior Engineer
SOX-compliant financial data platform
Delivered an encrypted data-transport and financial reporting platform with full audit traceability — moving sensitive financial data across systems under SOX controls, end to end.
