Technologist · Salt Lake City

From distributed systems to applied AI — years of building things that run.

I'm a technologist based in Salt Lake City. Today I build agentic AI for offensive cyber operations at Assail, and I'm scheduled to teach AI as adjunct faculty at Neumont University this fall. Most of my career has been spent on systems that have to hold up in production — distributed systems, connected products, real-time video, and applied AI going back to the early 2000s. I've worked as an architect, a CTO, and an individual contributor, and I still write production code. I've led and collaborated with engineering teams across seven countries. This page is a record of that work, and what I'm building now.

What I'm working on now.

Assail

Artificial Intelligence Engineer

Full-time · Aug 2026 – present · Boston, MA

Agentic AI for offensive cyber operations — autonomous red teaming at machine speed.

Neumont University

Adjunct Faculty, Artificial Intelligence

Part-time · Aug 2026 – present · Salt Lake City, UT · On-site

Teaching AI courses starting this fall.

Applied AI, decades before LLMs.

AI is one thread in a broader engineering career — but it is a long one, and it's the thread people ask about most. So here's the specific history. AI is having a gold-rush moment, and for a lot of people it's genuinely new. My own applied-AI work started just over two decades earlier: the AI Group at Countrywide Bank in the early 2000s, building expert systems for automated property valuation and loan appraisal. Followed in 2005 by a Guinness World Record for ChessBrain — a distributed chess engine running across 2,070 machines in 50+ countries. Chess is one of the fields AI used to measure itself against; I was building on that side of it. In the years since, the work has tracked the field as it widened. By 2013 I was doing natural language processing the hard way — PostAnalyzer, a Python and NLTK engine that pulled key phrases out of forum and social streams, scored their richness, and normalized slang, profanity, and misspellings into something a machine could reason over. That is the lineage a transformer architecture later turned into the LLMs everyone now reaches for — I was working the problem back when it still meant NLTK pipelines and hand-built feature rules. The thread kept widening from there, across modalities: real-time pose analysis in computer vision, speech-to-text transcription at scale, RAG pipelines, and LLM infrastructure. As CTO at Skafos AI, I led the team building an e-commerce product-discovery platform on vector image similarity — embeddings over product imagery, the same idea that now powers vector search and RAG.

The current chapter is agentic AI. At Assail I work on agentic systems for offensive cyber operations — autonomous red teaming at machine speed, which means giving agents real capability against real targets and keeping them accountable while they use it. Alongside that, I teach AI as adjunct faculty at Neumont University starting this fall.

Projects in the same thread: PNXStudios — AI-powered movement analysis from multi-angle video, for training, sports performance, and rehab; and Solrac — a self-hosted coding agent over Telegram that defaults to free local LLMs and escalates to Claude only on demand, with hard cost caps and an append-only audit log.

I'm also an active member of Utah's Silicon Slopes tech community, where I write regularly on AI, startups, and the craft of building. The pieces I'm proudest of are collected on my blog.

Four projects, in more detail.

Flywheel Sports → F45Chief Architect → VP of EngineeringConnected fitness · Live video · AWS · Redis · M&A

Directed 20 engineers building the Flywheel Sports tech stack, then helped sell it for $25M.

Flywheel Sports ran 42 brick-and-mortar cycling studios nationwide plus FlyAnywhere — a live and on-demand video platform streaming to connected bikes at home, in direct competition with Peloton. I directed 20 engineers building technology across the combined business: production studio automation, the video migration infrastructure (Verizon EdgeCast to Akamai during a live outage), and the systems tying the physical and digital sides together. FlyAnywhere ran on Hydra, the microservices framework I created that Flywheel later open-sourced.

FlyAnywhere was selected as one of Oprah's top picks for the 2019 Holiday Season. I subsequently supported the $25M sale of Flywheel Sports' technology to F45 Training.

F45 TrainingSVP of TechnologyPre-IPO · Kubernetes · API Gateway · Global engineering

Modernized the platform behind 1,500 studios in 60 countries, pre-IPO.

Brought on as F45's SVP of Technology & Head of Engineering to help prepare for F45's successful 2021 IPO on the NYSE — I was on the floor for the opening bell. Reporting to the CTO, I led engineering teams across six locations in four countries — New York, California, Texas, Sydney, the Philippines, and Vietnam — with about a dozen direct reports from Directors through Senior Architects.

Built F45's first AWS EKS Kubernetes cluster and led a platform-wide security upgrade — a new API Gateway and authentication service that now protects every one of the 1,500 studios. Created a container-first local development environment spanning multiple languages, platforms, and stacks.

Carlos provided me invaluable support in his role as SVP of Technology & Head of Engineering. He is highly regarded by our engineering team members and was a capable SVP of Technology and Head of Engineering. If afforded the opportunity I would absolutely hire and work with him again.

Dorian WorkmanFormer Chief Technology Officer, F45 Training — Carlos's direct manager
Fiveonefour (Moose)Principal EngineerRust · TypeScript · Open source · ClickHouse · Redis

Infrastructure that took an open-source framework to production at F45.

Over a two-year engagement, I shipped eighty-four merged PRs to Moose, an open-source analytical backend framework on ClickHouse and Redis. Built the coordination layer that lets multiple instances cooperate safely without stepping on each other — connection pooling, distributed locks, and leadership election for multi-instance deployments.

Spearheaded the initial effort to deploy Moose on Kubernetes in Google Cloud, taking it from single-box to a production-grade cluster deployment. Ran the security and reliability passes that hardened the framework for real-world use.

Also brought F45 Training in as Fiveonefour's first client — a relationship carried over from my time as F45's SVP of Technology — then joined the delivery team with key code-level contributions to the Lionheart 2.0 launch. The project rebuilt F45's analytics platform on Moose to serve 50K+ daily users and process over a billion biometric data points per week.

The rebuild boosted dev velocity for user-facing analytics by 10×, cut cloud infrastructure costs by more than half, and unlocked a year-long feature backlog. Downstream at F45: +70% member satisfaction, +44% app virality, and 2.5× higher lifetime value among members engaging with the new analytics.

PNXStudiosFounder & Principal EngineerAI pose estimation · Video · AWS · RAG · SOMA

A movement-learning platform built on AI pose estimation and 3D motion analysis.

Founded a B2B SaaS platform for movement-based practitioners — martial arts schools, yoga and Pilates studios, dance academies, and physical-therapy practices. Designed and shipped the platform end-to-end: mobile capture app, multi-camera 3D motion-capture pipeline, GPU-accelerated server-side media pipeline, multi-camera video player, 3D pose viewer, SaaS web app, and programmatic video editor. Built SOMA on top: AI-based coaching analysis of 3D movement.

PNXPlayer is a 29K-line TypeScript video player distributed via CDN and npm — a manager-based architecture spanning 28+ subsystems, with synchronized seek across stacked video elements, AirPlay and Chromecast, A/B loop, and transcript search with word-level timing. PNXCamera, the mobile counterpart for iOS and Android, synchronizes 4K capture across N devices via network channels with NTP-style time sync and audio-beep cross-correlation — sub-millisecond frame alignment, with network traffic in lobby state cut 77% via change-detection broadcasting. Photogrammetry-grade capture meant patching react-native-vision-camera at the native layer — Swift/AVFoundation on iOS, Kotlin/Camera2Interop on Android — for deterministic focus, exposure, and white-balance lock.

To manage the engineering work behind all of this, I built (a soon to be Open Sourced) agentic development workflow on the Claude Agent SDK with permission policy, per-chat cost caps, and multi-engine routing across Claude tiers and a local Ollama — anchored by a 450-document hybrid RAG (pgvector HNSW + Postgres tsvector, 70/30 semantic plus full-text with issue-ID exact-match boosting), exposed to LLM clients via custom MCP and UTCP servers.

People I've worked with, in their own words.

As a co-founder of Flywheel Sports, we prided ourselves in attracting and hiring an unusually passionate and hard working staff who cared for the business as if it were their own. Carlos is one of those people.
Ruth ZukermanCo-founder, SoulCycle and Flywheel Sports
Carlos is a unique individual who combines a rare capability to look at a problem as a whole, at high level, and can also break it down to the smallest excruciating detail. He finds common language with senior executives who don't understand technology, while having complete respect of his developer team. At Flywheel he served as VP of Engineering, but also acted in the capacity of the CTO.
Elena SukachevaCEO, UCAN — formerly managed Carlos at Flywheel
Carlos and I worked together at Skafos AI while he was our CTO. Under his leadership, our team onboarded several new engineers, undertook a significant architectural shift, and grew our customer base from 0 to several launch partners in a few months.
Tyler HutchersonApplied AI @ Redis — formerly direct report at Skafos AI
When an external enterprise vendor nearly wiped out all of Flywheel's online videos, Carlos developed a replacement service and migrated the missing files to an alternative cloud solution within five days. Absolutely incredible.
Greg SolakHead of Engineering, F45 — reported to Carlos at Flywheel and F45

The rest of it, in shorthand.

Guinness 2005

First distributed network in history to play against an internationally ranked human opponent in a public exhibition — ChessBrain, 2,070 machines across 50+ countries.

5 exits

Companies I worked at that went on to an IPO or acquisition: F45 (2021 NYSE IPO), Flywheel (technology sold to F45), Estate Assist (DocuSign), IGT (GTECH), inSparq (Adiant).

Earlier work

Earlier in my career I worked on projects for Activision, Disney Interactive, IBM, and Symantec.

Oprah ’19

FlyAnywhere — Flywheel Sports' connected-fitness platform — was selected as one of Oprah's top picks for the 2019 Holiday Season.

25+ years

Applied AI since the early 2000s: the AI Group at Countrywide Bank, distributed AI on ChessBrain, early NLP with NLTK, vector image similarity at Skafos AI, today's RAG and LLM infrastructure.

Books

Co-authored Advanced Macromedia Flash MX ActionScript in Action (Prentice Hall, 2002) and the ChessBrain chapter in Advanced Intelligent Paradigms in Computer Games (Springer, 2007). Peer-reviewed work at IEEE CIG and ICGA Journal; writing in Linux Journal, IBM developerWorks, and O'Reilly. CIO Review (2019), “Evolution of Wearables and Connected Fitness.”

Talks

RedisConf 2018, 2019, 2020 — the 2020 talk on BioGraph, a human-physiology model on RedisGraph, was selected as a best talk. EmpireNode 2016. BoF Linux Conference Copenhagen 2005, keynote.

10K+

GitHub commits since 2010. Maintainer of Hydra and other open-source projects.

12+ shipped

Twelve-plus commercial products across the career — from business applications and computer games to distributed AI, connected fitness, and movement analysis.

ACM · IEEE

Member, Association for Computing Machinery and Institute of Electrical and Electronics Engineers. Member and contributor, Silicon Slopes Technical Community, where I write on AI and startups. Redis Stars honoree.

Hardware

Custom PCB design and manufacturing, ESP32 / Arduino / Raspberry Pi firmware, BLE, 9-DOF IMU/AHRS, biometric sensors, PlatformIO. The physical-product side of connected products.

3 → 17

Director of Software Development at iPayment, Inc. — grew engineering from 3 to 17 as company revenue moved from $50M (2010) to $70M (2011). iPayment was later acquired by Paysafe Group in 2018.

Get in touch

If you'd like to reach out.

Email is the best way to reach me.

cjus@ieee.org