Vasiliy Uvarov — Work record: roles, side projects, and verified metrics.

As Chief Data & AI Officer at Altery, Vasiliy Uvarov builds and operates internal agent systems inside a UK electronic-money institution: systems that know where company context lives, call the right tools under real permissions, leave an audit trail a compliance officer will accept, and run at a cost that survives a budget review.

The FCA-approved EMD Individual standing is separate from that AI work. It concerns management of an electronic-money institution generally; it is not an AI qualification or an AI-specific accountability regime.

The platform layer. 230+ production-grade skills across 14 domains — compliance, engineering, marketing, data, QA, operations, agents, documentation, product, startup, document processing, developer tools, AI coding agents, and Altery regulatory — dispatched by an 8-router orchestration system. Routing is deterministic and auditable rather than model intuition: hand-authored routing maps, a machine-readable registry with explicit confidence thresholds, and a drift gate that fails the build when the checked-in skills graph differs from a fresh export. The library is mirrored publicly under MIT, so the structure can be inspected without an NDA.

The coordination layer. requirements-hub— a cross-repo coordination repo that makes a large service estate discoverable to humans and agents alike. Deterministic scanners profile each service’s APIs, databases and infrastructure into normalised records; deployment state is read from the GitOps source of truth rather than from memory; repo-local documentation coverage is tracked as a governance KPI instead of left to goodwill; and a platform-wide gap analysis holds 545 gaps categorised and remediation-tracked across 12 top-level domains.

The operator layer. Enterprise-wide AI platform with governance across 7 departments — 20+ Custom GPTs and 9 Agent SDK apps in production, with near-universal engineering adoption. The flagship agents ship as services, not prompts: a customer-support service that replaced a 78-node no-code workflow, with OpenTelemetry, Prometheus, vaulted secrets and PII scrubbing, and a config-driven fraud and PII detection engine with parallel-versus-single-pass equivalence tests. Both suites green.

Built the product and platform foundation for a regulated fintech and crypto-infrastructure business, and scaled it to €8-figure monthly GMV. Led Visa and UnionPay principal membership work, in-house clearing, multi-currency cards (GBP/USD/EUR), Apple Pay / Google Pay tokenisation, crypto-to-fiat ramps (TON / USDT / NOT) under 60 seconds, paperless proof-of-address in 29+ countries, and payout rails across 10+ markets.

Stood up the analytics and operating cadence from a blank page — a self-serve BI migration, an OKR cycle, seven cross-functional squads, and monthly company-wide demo days. CAC per transacting user fell roughly 2.5× over the period.

Partnered with an executive team to rebuild product and growth systems: UX audits and funnel tuning that roughly tripled onboarding conversion, and scoping discipline plus PM coaching that cut the feature backlog by more than half. Left behind a product-ops model — an experimentation pipeline, OKRs embedded in roadmap planning, and an org design spanning product, engineering and compliance — rather than a set of recommendations.

The client, its sector and its scale are covered by the engagement NDA. The underlying figures are available in a conversation under the same terms.

Owned onboarding and the merchant lifecycle for an EU and international B2B payments platform. Cut KYB onboarding from ninety days to twelve through automation and process redesign, and improved merchant time-to-revenue by seventy percent by simplifying the funnel. Introduced OKRs tied to activation, drop-off and LTV; backend optimisation cut transaction costs by 52%.

Led onboarding and master-data architecture for financial institutions and broker platforms. Replaced fifty percent of legacy dependencies — unlocking £1.7 million in annual savings and a thirty-five percent cut in onboarding abandonment. Rebuilt KYC/AML as modular, region-ready flows.

This role overlapped with the Taiger role below. Both date ranges are accurate: they ran concurrently during a relocation from Moscow to Singapore.

Owned solution design and go-to-market for AI-powered KYB and document-extraction tools. Closed £321K+ in Tier-1 bank deals through demo-led enterprise cycles — discovery, technical evaluation, deployment — and drove product-market fit for OCR/NLP compliance modules holding 90%+ recognition accuracy. Built a 25-plus lead pipeline in under six months and launched a bundled go-to-market with delivery partners.

The only enterprise-AI role before Altery, and the closest earlier analogue to the current one: scoping a document-extraction problem with a bank’s compliance team, then proving accuracy against their own documents.

Built and scaled an SME lending platform from ₽70 billion to ₽130 billion; cut origination time from fourteen days to under one via low-code orchestration. Deployed a hundred-plus RPA bots across SME onboarding and operations, optimising ninety-plus FTEs of process labour. Shipped AI-powered KYB with 92–94% OCR accuracy.

Oversaw a £100 million-plus IT portfolio across 21 teams and 230-plus engineers — CRM, onboarding, mobile, lending, and core systems. Prepared financial models and KPI projections for bank committees.

This role ended in October 2021, before the February 2022 sanctions regime.

Astrology and divination platform. Native iOS in Swift plus a Next.js web app, sharing a backend, localised into fourteen languages. Chart calculation runs on a deterministic VSOP87 ephemeris held to ±0.1° against JPL Horizons — not an LLM guessing at planetary positions. The build receipt covers the cosmic copilot native app + web app and is dated 21 May 2026.

Hallucination detection, citation-drift tracking, competitive intelligence, and recommendations for getting cited by answer engines. The eval discipline is the product: every claim an engine makes about a brand is graded against a source.

A Python pipeline ingests and normalises apartment listings; a Next.js and Firebase front end publishes them, tuned for answer-engine and search visibility. The point-in-time receipt covers the prime legacy homes repository and was measured 21 May 2026.