Independent builder · AI-agent systems
Quant and trading infrastructure, resilient data pipelines, monitoring, and automated security review — designed, shipped, and operated by one person orchestrating fleets of AI agents. Senior output, a fraction of the cost and headcount.
Capabilities
Six things I have built and operate in production today — not a wish list. Each is available as a standalone engagement or as part of a larger system.
Fleets of headless agents that research, build, and verify technical work on a schedule — with an adversarial check step so nothing ships unreviewed. The reason one person can run all of the below.
A dozen-plus live collectors (exchange APIs, on-chain, options/vol, SEC filings, market reference) with atomic writes, retention/rotation, and freshness monitoring. They don't silently die.
A harness that refuses to fool itself: point-in-time, survivorship-honest panels, walk-forward splits, placebo/permutation controls, realistic cost models. It tells you when an edge is noise.
A static-audit pipeline that reads source the way a reviewer does — two-track (invariant sweep + open search), scope- and duplicate-gated, with an adversarial verification pass before anything is reported. No scanner noise, no fabricated findings. A broken-authorization bug I reported to MLflow is fixed and merged upstream — PR #25721.
Order and stop management across Binance, KuCoin, and Solana, with position state, exchange-truth reconciliation, and fail-closed safety gates. Shadow mode before a cent moves.
Watchers that track filings, prices, and on-chain events and message you only when something real happens — built so a failed fetch alarms as "blind," never as "nothing new."
Proof of work
Over the past year I designed, built, and have continuously operated a production quant-and-security platform on a single server. It runs sixty-plus scheduled jobs: data collection, research, live execution, and self-monitoring — the whole loop, unattended.
What makes it unusual isn't size, it's discipline. The research layer is built to disprove its own ideas, not confirm them. The monitoring layer distinguishes a dead sensor from a quiet one. The security layer only reports a finding once it survives an adversarial second pass. That is the same judgment I bring to a client's codebase.
Shown as an anonymized capability reference. I can walk through the architecture live, under NDA where needed.
Why you can trust the output
The reason this isn't another AI-generated firehose: nothing counts until it's checked. That discipline is the actual product.
Nothing is reported until it survives a second, hostile review. A finding that can't be reproduced doesn't ship — no scanner noise, no fabricated results.
Each audit runs an invariant sweep and an open search in parallel. They have different blind spots, so together they catch what either alone would miss.
Research is built to disprove its own ideas — walk-forward, out-of-sample, placebo. I'd rather tell you an edge is noise than sell you a dashboard.
Engagements
Fixed-scope projects with a clear deliverable and timeline. Prices are starting points — the real number depends on scope, which we set together in a short call.
Ongoing or retainer work runs at a day rate of $600–900. Based in Switzerland; work remote worldwide, comfortable across EU/US hours.
Why the economics work
I do the design, judgment, and review myself — and hand the repetitive build-and-check work to fleets of AI agents I orchestrate. That's how one person ships and runs what normally takes a small team.
You get the quality of a senior engineer who has actually built and operated this kind of system, at a rate that reflects the leverage, not a big shop's bench. You pay for the thinking; the typing is nearly free.
The standard
Swiss-built — precision, discipline, control. Raised between cultures — Swiss structure, Italian fire. My son comes first. That's my standard.
— Dev of ShelbyCore
Get in touch
A short description of the problem is enough to start. I'll tell you honestly whether it's a fit and roughly what it takes.