infrastructure architect · ai builder
I architect the infrastructure that runs trading systems — and I build AI with the same discipline.
I'm Darrell — a platform & trading-systems architect, and a lifelong builder of things that actually have to work. This site provides a view into some of the AI platforms I've developed recently, as well as some of the other fun projects I'm enjoying. If you look around a bit, you'll notice a common thread: the model is a co-pilot, never the autopilot — they reason, they propose, they draft, while deterministic gates make the real decisions that you can't take back. The rest of it — the hardware, the hand-built synths — they share a different set of rules: always finish what you start and have fun along the way. Please use the contact form to leave a comment or to make me aware of a collaboration opportunity.
// selected work
Case studies
Devy
Agentic DevOps · 2026
An open-source agentic DevOps/SRE assistant — one capable agent, an on-demand tools-router, hybrid-RAG citations, a no-shell host boundary, and timeline-based incident root-cause analysis. The reusable, from-scratch rebuild of a co-pilot I first built for a production trading platform.
PreDEX
Agentic AI · 2026
A DEX perps-trading platform whose Claude-powered co-pilot proposes, scouts, and critiques across the whole strategy lifecycle — but can never execute a trade.
PredMark
Agentic AI · 2026
Monitors ~101K Kalshi and Polymarket markets for edge, trades through deterministic gates, and uses Claude for research and measurement — but never for the trade decision itself.
Luna
In-App Agent · 2026
A complete, end-to-end commerce platform for a growing hand-crafted jewelry business — products down to their components, suppliers, and multi-channel sales — fronted by Luna, an AI that takes the toil out of running it.
Lumen Atelier
Autonomous Creative AI · 2026
A near-open-ended commission: build something visually dazzling that uses AI in a way that hasn't been done to death. The result is a studio that renders its own work, critiques the rendered frames with a vision model, and revises until a piece is gallery-worthy — every finished piece shipping as live shader code that runs on the visitor's own GPU.
// lab
Experiments & fun
Here's a collection of some smaller projects that I've enjoyed working on. Many are available as open source software (OSS) if you'd like to play around or take a crack at improving them. In stark contrast to modern AI systems, you'll notice that I definitely have a thing for 80's retro and hardware from that era.
CubeLab
A photoreal, GPU-accelerated 3D cube with standard notation, mouse and keyboard controls, repeatable scrambles for timing your solves, and a near-optimal Kociemba solver that finishes it for you when you're stuck.
LLM on a Raspberry Pi
A fully dockerized, open-source path to a local LLM on a Raspberry Pi 5 + Hailo NPU — Ollama under the hood, a clean Open WebUI on top, and a genuinely capable assistant running entirely on-device.
Vision Pipeline
A self-hosted, real-time vision pipeline: one camera feed, fanned out over a Redis bus to as many models as you like — YOLO boxes and a chat-driven Moondream VLM today, a perception foundation for future robotics. Runs on Apple Silicon or a Jetson.
Ambika
A scratch-built, fully-loaded Ambika — six analog voicecards in two filter flavors, seven hand-flashed AVRs, a custom wood-cheeked case, and a 700-patch library. A deep DIY build of Emilie Gillet's (Mutable Instruments) design.
EVO64
A premium, surface-mount C64 mainboard that folds the community's best mods into one integrated board — supported by a GPT-4o Discord assistant that answers builders' questions 24/7.
MB-6582
A hand-built MidiBox SID synth: eight rare MOS SID chips, four PIC microcontrollers, a custom control surface, and a lot of soldering. A deep DIY build of Wilba's MB-6582 on Thorsten Klose's MidiBox platform.
// about
For over two decades I've built and led the infrastructure behind electronic trading platforms and low-latency systems, where a bad deployment is never an option. Right now I'm the Global Head of Trading Systems Infrastructure & DevOps at FalconX, where I had the opportunity to design a greenfield low-latency trading platform architecture from scratch — it cleared ~$2B+ in notional flow within six months of its launch.
The platform makes pervasive use of AI throughout every stage — from live agents that support the trade blotter and market-data ticker-plant, to MCPs that support and manage trading execution hosts, through to a deep knowledge-base retrieval system that knows the architecture of every component, from their active operational states, down through the source code.
For obvious reasons I can't share the details of that proprietary platform here, but I've recently developed a brand-new Agentic DevOps platform as a personal project, which I'm happy to make open source under the Apache license. Check out Devy in the Case Studies section.
If you're building with AI, just remember that the hardest parts live in the pipeline. Get the agent harness, the guardrails and the observability right — that will earn the user's trust.
Prior to my current role, I was Chief Infrastructure Architect for a multi-billion-dollar fintech, where I designed their ultra-low-latency compute and networking systems for extreme-velocity fractional trading. Prior to that, I ran global electronic-trading infrastructure and high-performance hosting for Credit Suisse.
When I'm away from work, I never stop tinkering. The lab here features hand-built analog synthesizers and a ground-up, complete reimagining of the old Commodore 64 from the 80s — that's the machine that started all of this when I was 14. A different scale and a different time, but the same instinct: pick something worth doing, get it all the way to done, and bring others along for the ride.
what I reach for
- ▸ TypeScript · Python · Go
- ▸ Next.js · FastAPI · Postgres
- ▸ Anthropic · OpenAI · pgvector
- ▸ Railway · Docker · CI/CD
// contact
Building something where the details matter?
I'd love to hear about it — whether it's a role, a collaboration, or just to compare notes.
A quick verified sign-in keeps the inbox spam-free — nothing is shared.