Currently · Blits.ai · Amsterdam

Agentic AI,
shipped
end to end.

AI Product & Delivery Lead — owning agentic AI from problem to production: multi-LLM orchestration, evaluation-based governance, and pre-production risk sign-off for Tier-1 financial institutions.

Before Blits.ai: Team Lead at kimkim ($4M+ ARR, 8 destinations), founder of Canicula.club, and Lead Analyst at the Office of the President of Ukraine.

02 / Proof
10yrs Launching advanced tech in production
6 AI engagements closed with Tier-1 banks
4 Digital humans shipped end-to-end at Blits.ai
5 Countries covered by banking AI delivery
$4M+ Annual revenue at kimkim · 3× bookings growth
60% Manual workflows removed via API integrations

What the role asks for. What the experience shows.

The role asks for

My experience shows

Drive end-to-end delivery of LLM and agentic AI programs across engineering, data, risk, and business stakeholders.

Currently owns technical delivery of conversational AI and digital human products at Blits.ai.

Translate ambiguous AI capabilities into scoped roadmaps, milestones, owners, and delivery rhythms.

Leads strategic AI partnerships with Tier 1 financial institutions across the Middle East, Europe, and Turkey.

Own model evaluation, pre-production risk review, and executive sign-off for regulated use cases.

Produces vulnerability assessments and LLM risk reports that support go/no-go decisions.

Coordinate vendors, platform APIs, STT/TTS providers, data pipelines, and inference infrastructure.

Architects multi-LLM systems with conversational and tool-use layers, including agentic payment flows.

Keep teams unblocked while surfacing risk early to banking project managers and sponsors.

Drives build/buy/partner decisions across frontier models, inference, data, STT, and client platform APIs.

Turn pilots into repeatable delivery playbooks.

Leads Blits.ai's LLMs4EU evaluation workstream, defining metrics and methodologies with European partners.

The role asks for

My experience shows

Define the problem and the user before the spec — turn ambiguous market signals into a product thesis.

Defined kimkim's market-entry approach for new destinations — which markets to open and why — before building.

Set success metrics up front, then read post-launch signal to decide what ships next.

Set metrics and read the signal: removed 60% of manual workflows and grew bookings 3× after shipping integrations and activity scoring.

Make product decisions under ambiguity — scope, sequencing, and stack choice without complete information.

Chose the AI stack under ambiguity — model, inference, STT/TTS, and data — through build/buy/partner calls and A/B testing.

Treat governance as product surface — evaluation, guardrails, and sign-off users can trust.

Built the evaluation suite as a product with users: defined metrics and methodologies via LLMs4EU that gate go/no-go.

Convert discovery into a prioritized roadmap and OKRs across engineering, data, and commercial.

Set strategic roadmaps and OKRs at kimkim, analyzing results to prioritize what to build next.

Own the product end to end, from problem definition to shipped, monitored production.

Shipped agentic flows end to end at Blits.ai — payment, catalog search, and tool-use across a multi-LLM stack.

The role asks for

My experience shows

Set AI product direction across roadmap, architecture, partnerships, governance, and commercial value.

Owns build/buy/partner decisions across model selection, inference, data pipelines, STT, and integrations.

Make build/buy/partner calls across the agentic AI stack and vendor ecosystem.

Provides forward-looking product strategy on dialect NLP gaps, avatar inference cost, and agentic architecture trade-offs.

Define product evaluation, risk posture, and adoption path for regulated customers.

Turns banking requirements into scalable partnership frameworks and repeatable AI delivery playbooks.

Shape client discovery into product requirements, reusable patterns, and delivery strategy.

Architects product flows across payment APIs, CRM, core banking systems, and real-time merchant integrations.

Represent AI trade-offs clearly to executives, technical teams, and commercial stakeholders.

Defines LLM evaluation benchmarks and risk methodologies through LLMs4EU.

Scale product ideas from pilots into partnership frameworks.

Previously owned roadmaps, OKRs, product evolution, content strategy, UX analytics, and commercial growth at kimkim.

Resume evidence.

05/2025 — Present

Technical Program Manager

Blits.ai · Amsterdam

Currently leading delivery of conversational AI and digital human products for six Tier-1 banks across MEA, Europe, and Turkey.

  • Drove build/buy/partner decisions across model selection, evaluation, inference, data, STT, and embedded APIs.
  • Architected multi-LLM orchestration with conversational and tool-use layers — including agentic payment flows and SQL-driven catalog search.
  • Aligned client executives, product, engineering, and compliance teams around AI capabilities and delivery risk.
  • Delivered vulnerability assessments and LLM risk reports for go/no-go decisions.
  • Led the LLMs4EU evaluation workstream — defining benchmarks and methodologies across European partners.
05/2023 — 05/2025

Team Lead

kimkim · Amsterdam

Generated $4M+ annually across 8 destinations by leading cross-functional teams scaling AI-driven travel-tech.

  • Defined strategic roadmaps and OKRs; analyzed results to prioritize AI product evolution.
  • Spearheaded OpenAI integrations and scoring algorithms for travel activities.
  • Integrated Bokun and FareHarbor APIs — reduced manual workflows by 60%.
  • Improved UX flows through FullStory analytics and shadow sessions.
  • Owned content strategy and production of 500+ pieces annually to grow leads and engagement.
  • Improved supplier relations and in-trip support — 3× bookings, ⅓ issues.
05/2022 — 05/2023

Destination Development Manager

kimkim · Amsterdam

Launched a full-scale travel-tech product from scratch and managed multiple pivots within a year.

  • Built development and operational frameworks for Greece and Costa Rica.
  • Managed lead-generating content creation with a team of 3 editors.
05/2012 — 05/2022

Owner & Founder

Canicula.club · Remote EU

Built a custom platform for long-duration adventure tours serving 500+ clients with 50% annual growth over five years.

  • Negotiated supplier partnerships and beneficial B2B/B2C terms.
  • Implemented Google and Facebook ads, optimizing ROI through data-driven testing.
01/2008 — 05/2012

Lead Analyst

Office of the President of Ukraine · Kyiv

Led large-scale data analysis and financial modeling informing government decision-makers.

  • Developed and refined macroeconomic models (Excel, VBA, SPSS) for policy decision-makers.
  • Analyzed socio-economic and reform policy effectiveness; supported change-management frameworks.

Technical stack

SQL, Looker, Power BI, A/B testing, Azure DevOps, Jira, Confluence, Notion, Figma, Scrum, Kanban, LLM orchestration, TTS/STT APIs, prompt engineering, LangChain, RAG, Python, VBA, C, digital human engineering.

Product leadership

Product design, architecture, requirements, stakeholder leadership, operational optimization, SEO, content strategy, team-load projections, roadmap ownership, and delivery governance.

Education

M.Sc. Computer Systems Analysis · Kyiv Polytechnic Institute. B.Sc. Mathematics & Computer Science · Kyiv Polytechnic Institute.

Languages

English fluent, Ukrainian native, Dutch A2, Spanish limited working.

Public projects.

Retirement Planner NL

Helps expats plan retirement in the Netherlands. Live web app, built and maintained solo — backed by three weeks of in-depth research.

Open the app

No-Fluff Media Converter

Shrinks screen recordings to email-size without visible quality loss. Quality- or size-targeted modes, side-by-side preview.

macOS · Swift + ffmpeg

Transcriber

On-device speech intelligence: record, transcribe, diarize, translate — nothing leaves the phone. Apache-2.0, sideload APK.

Android · Gemma 4 + Whisper.cpp

OpenScan

Document scanner with ML Kit edge detection and perspective correction; no camera permission required. Exports multi-page PDFs.

Android · Kotlin + Compose

DiscStats

Disk-usage treemap in a single SwiftUI Canvas, with hard-link and APFS firmlink dedup so totals match Finder.

macOS · Swift

RainLock

Touch-blocking overlay so rain or curious fingers can't trigger taps; unlocks with a volume-button combo.

Android · Utility

Talk to me directly.

CV Download PDF Email Click to reveal Phone Click to reveal LinkedIn ihnatov-yuri
Location Amsterdam, NL