I'm Brayden Hord. I run engineering, IT, data, and AI at Arena and sit on its executive team. I run the function like a P&L: every system my team ships carries a value model, the model gets audited, and nothing counts until real people are using it. Personal projects are at the bottom of this page.
A small group of engineers who own their work end to end, from spec to on-call. Twenty-five systems in production. The rule is simple: it isn't done until it runs for real users, is valued honestly, and the people it was built for actually use it.
25 systems live · 4 AI models in productionI wrote the enterprise AI policy, chair the AI and Automation committees, chose and rolled out Claude Enterprise and OpenAI Enterprise for 100+ staff, run the training, and ship the agents myself: Claude on AWS Bedrock in production, and local models (CLIP, Ollama) wherever the data can't leave the building.
Claude · OpenAI · Bedrock · local modelsIAM across AWS, GCP and Google Workspace, hardware tokens, quarterly security training, a vendor-assessment rubric (SOC 2, ISO 27001, breach notification, training-data opt-out), and a 24-hour incident protocol. Boring on purpose.
AWS · GCP · Workspace · hardware keysA value model across 49 inputs in nine categories sits behind every system: revenue protected, retention, media efficiency, errors prevented, billing accuracy, risk avoided, labour returned, cost avoided, new revenue. COGS, margin and EBITDA per build. When an audit says a headline number is too high, it comes down.
$1.2M realized · every dollar a formulaAbout 175 staff hours a week returned, about 4.4 FTE. A daily compliance sweep across Meta, StackAdapt, and X accounts catches committed media running dark. A four-hour reconciliation loop across eight platforms that flags money moving with no sold line behind it. A build-vs-buy that replaced a $50K/yr vendor feature with a script.
~175 hrs/week · 2,000+ accounts onboarded per cycleTwo productized offerings shipped, an internal tool turned into a metered, billable service, $250K+ in technical consulting sold from a sales-engineer seat, and a proprietary dataset that gets more valuable every month it accrues. Client-driven and revenue-bound, or it doesn't get built.
2 products shipped · $250K+ consulting soldEvery system my team ships carries a live financial model. Value is modeled across 49 inputs in nine categories, and every dollar is a formula pointed at auditable inputs, each rated for confidence. No asserted totals, anywhere. A figure you can defend in front of a CFO beats a bigger one you can't.
That discipline changes what gets built. Builds are weighed the way finance weighs them: what they do to COGS, what margin they protect, what they contribute to EBITDA. Some have grown into intellectual property with standalone value, like a metered service that turned an internal tool into a billable product line, and a growing proprietary dataset that compounds as it accrues. A few of the quieter examples: the daily compliance sweep, the reconciliation loop, and a privacy toolkit that lets client data be used with third-party AI without handing the data over.
None of it is a solo act. I lead a tight-knit team of engineers who own their systems end to end, and the operating rule is that nothing counts until it's running in production for real users, valued honestly, and adopted by the people it serves.
Shows what every ad pixel on a page sends, proves whether tracking stops when a visitor says no, follows an ad click through to the purchase, and tests new pixel code in a Safe mode that never sends a real event. Made for anyone who ships tracking and wants proof it behaves.
Sends a synthetic visitor through a site, has her say no, and runs every tag for real inside a browser that can't reach the internet. Then it shows which vendors tried to collect her data anyway, who stored an ID for later, and who recognized her after she withdrew consent. Nothing ever reaches a vendor.
Watches the header-bidding auction happen in your browser: every Prebid bidder and price, including the bids that lost, the creatives behind them, the IDs you've been assigned, and the targeting that was sent. Then it lets you change that targeting and prove a line item serves.
Keeps a MacBook awake with the lid closed and no monitor attached, then turns itself off the moment you unplug, so it never cooks in a bag. One obscure system flag plus the guardrails that flag badly needs. Swift, no dependencies, free.
Hand it the email thread and the project board and it places the media buy across DSPs and social platforms, writing every action back to the PM system so nothing happens off the record. Built on Claude through AWS Bedrock with 36 custom platform tools. What it changed: a buy that took a planner an afternoon of copy and paste now takes a review and a click, and the audit trail is better than the manual one was.
A search engine for a library of thousands of mail pieces and digital ads. Describe what you need in plain English and it returns the closest matches, ranked by a CLIP model that runs on the box itself, so no creative ever leaves the network. Facet filters, PDF rendering, ZIP and slide-deck export, and a usage monitor so adoption is a number, not a feeling. Runs on a spare Mac on the LAN or a $22-a-month cloud box.
A privacy layer that lets proprietary data be used with third-party AI tools without handing the data over: records are pseudonymized, totals stay exact, and the guarantees are documented in plain terms for leadership and for client security questionnaires. The point: use the vendors without leaking the records or the aggregates.
A naming system for ad objects, live across StackAdapt, Meta, LinkedIn, TikTok, and Google, designed so reports join cleanly across platforms instead of being reconciled by hand.
Closes the loop between a mail drop and what happens online. USPS Informed Visibility scan ingestion, a Flask PURL system for personalized landing pages, and a research portfolio across BLE beacons, conductive ink, and Digimarc. Physical spend finally shows up in the same dashboard as digital.
One normalized schema and versioned spec for client-services to media handoffs across Meta, Programmatic, YouTube, Search, Native, X, LinkedIn, and Reddit. Recovered 62 team hours a week and ended a decade of tribal-knowledge drift.
A QA bot in Slack: the Claude API behind Make.com, tuned to hold a QA standard across media ops without being annoying about it. It catches the small mistakes people miss, and the team moves faster because someone is always checking.
Three ESP32 nodes talking over ESP-NOW, HC-SR04 ultrasonics, and a Waveshare 64×64 HUB75E LED matrix with a weather display and pixel-art animations. Timing-sensitive matrix driving and sensor fusion across nodes. A personal build, made after hours.
A password generator that is honest about its randomness, a
Fisher-Yates shuffler, Math.random() plotted next to
32-bit xorshift, a NIST-style RNG testing kit, an F1 reaction
timer, and Elemental Arena, a bouncing-orb battle game where
every match replays exactly from a seed.
I lead engineering, IT, AI, and innovation at Arena (arenawins.com), a digital advertising firm with deep political-cycle exposure and an unusually broad surface area: DSPs, social, CTV, direct mail. The role was created on purpose, as a distinct vector for engineering next to traditional agency work. That is my day job, and the ArenaWins link in the header goes there. Elemental Arena, elsewhere on this site, is a game about orbs and has nothing to do with it.
I'm most useful where three things meet: production AI that has to actually run, hard data and information-theory problems, and the human work of getting a whole organization to adopt new machinery. I own the AI strategy end to end. I authored our enterprise AI Policy v2.0 (7 core principles, 16-category Universal Data Restrictions, a vendor-assessment rubric, a 24-hour incident protocol), chair the AI and Automation committees, deploy the agents myself, secured, governed, and monitored, and run the training that gets 100+ people using them.
Outside work I build embedded electronics, run statistical tests on PRNGs because it's fun, contribute to local political campaigns, and keep a homelab in northern Utah running on more YAML than is strictly defensible.
Technology leadership conversations welcome — VP Technology, CTO, and Head of AI / Engineering roles at 50–500-person companies. braydenhord@gmail.com
