AI systems that do expert work — shipped.
Independent architect designing multi-agent AI systems that research, build, adversarially verify, and ship real products.
Four practice areas, one operating principle: systems that are verified before they are trusted.
Multi-agent swarms, model routing across vendors, verification pipelines, and memory systems that let agents accumulate context safely over months of operation.
Research and engineering of systematic trading infrastructure: data pipelines, backtest harnesses, execution plumbing, risk controls.
Engineering only. No performance promises. No investment advice.
Decision-support tools grounded in primary UK law, with citation-verified pipelines that check every authority against live legislation before it is used.
Fixed-scope engagements: architecture reviews, agent-stack builds, and automation audits. Defined deliverable, defined price, defined end date.
Not one model answering a prompt — a supervised production line.
An orchestrating agent decomposes the problem and dispatches specialist agents — research, drafting, engineering — each on the cheapest model capable of the job.
Separate audit agents — different models, different context — attack the output: factual claims, citations, edge cases, failure modes.
Nothing ships on machine consensus alone. A human reviews the audited output, resolves disputes, and takes editorial responsibility.
Corrected, verified, and released — as a product, a codebase, or a report. In public, with a paper trail.
Three UK legal self-help products — written, adversarially audited against live legislation, corrected, and launched in a single night by an orchestrated AI swarm under human editorial control.
Step-by-step self-help pack for bringing a UK Employment Tribunal claim.
UK immigration self-help guidance, grounded in primary legislation.
Navigating a UK workplace disciplinary process, procedure by procedure.
Fixed-scope consulting engagements and product collaborations. One email, a short scoping call, a written proposal.
Zitanvault@proton.me