Vlad Derkach
EN RU PL
Verify →

Beat 01 Scene

You're reading the fortieth product manager's CV this week.

Every one says "AI-first." Every one carries numbers nobody can verify.

I'm Vlad Derkach. I was Product Discovery Lead and AI Champion at RedTrack.io, a MarTech attribution and analytics platform. Five years in product, twelve building businesses before and around it.

A mis-hire for an AI lead costs six months and a budget. So I won't try to convince you. Below is what I've built — and where you can inspect it without any help from me.

Start with the first exhibit Skip to verification

Open to Head of AI Product, AI Product Lead, and Principal PM (AI / agentic products) roles — B2B SaaS, scale-ups and Series A/B. Remote or Gdańsk / Tricity. Plus one or two consulting engagements at a time.

Vlad Derkach
25+
agents in production
17
competitors monitored weekly, automatically
$2.3M
smart-contract volume · Roketo
33 → 75%
onboarding completion · FemFast

For now, these are four numbers on a page. What stands behind them comes next.

Beat 02 Exhibit A

What's running right now

Built at RedTrack, Nov 2025 – Sep 2026 · still running there.

In 2026, "we've adopted AI" usually means a demo on an internal call and a slide in the quarterly report.

I joined RedTrack in November 2025. Discovery lived as a project phase — ad hoc. No AI ran in operations. Meanwhile the company was shifting from an affiliate-heavy customer base toward e-commerce and DTC, with a competitive migration window closing on top.

I turned Discovery from a phase into a company function. A Discovery Request standard now sits at the front of every hypothesis — teams adopted it without a single edit. Behind it: a live hypothesis backlog and a Discovery → decision → delivery flow that produces C-level risk memos before big bets.

Around that function, 25+ Claude agents run in production. Each one has a schedule, logs, and a place where its output lands.

live

Competitor Scout

Weekly intelligence across 17 tracked platforms — changelogs, pricing, positioning — synthesised and delivered to Slack.

live

Support AI Assistant

Reads each customer's live account settings behind two PII layers, not just the knowledge base. Built with my first direct report.

live

MCP Skills Pack

Delivers value inside users' own AI agents before they pay — B2A infrastructure acting as an acquisition channel.

11
scheduled routines
25+
production agents
17
platforms monitored weekly
Thousands
of support tickets analysed

What changed. Four months in, I entered the Leadership Hub — the C-level strategy body. The mobile app shipped on iOS and Android. E-com and affiliate dashboards were defined, built, and released. MCP went live in production in June 2026. Twelve strategic outputs were personally green-lit by the CEO over the first seven months.

All of it lives inside the company, and I can't hand you a password. The next three exhibits sit out in the open.

Beat 03 Exhibit B

What happened when things broke

Numbers on a CV are fair-weather results. What matters is what a person does in a storm.

Roketo · Product Manager → Chief Product Officer · May 2022 – August 2023. Six frames, one per turn.

01 The hero

May 2022. I'm the PM

Roketo streams tokens as smart contracts on NEAR. In my first six months I rebuild the on-chain flow: conversion ×4, $2.3M moved across 7,600 transactions, partner reach ×3 through API integrations.

02 The problem

The market turns

Web3 slides into a down-cycle. External funding closes. Token-streaming as a category narrows to one living use case — a wallet with payments.

03 "Oh no"

The holding cuts half

Between Q3 2022 and Q1 2023, half of Kikimora Labs — the parent holding — shuts down. There will be no round.

04 The move

I'm made CPO

Promoted from PM to Chief Product Officer to lead the response. I redefine the product as a wallet with an integrated payment solution and relaunch Roketo as QUV.io.

05 "Aha"

$70K in grants

The runway money arrives as grants. No round — and the founders' equity stays intact.

06 After

The team holds

Everyone makes it through the pivot — in the same period the holding lost half its projects.

The tally

$2.3M smart-contract volume in 6 months · 7,600 on-chain transactions · ×4 conversion · ×3 partner reach · $70K grants secured · zero team losses while the parent holding lost half its projects

FemFast: an MVP in two months, in a category Meta restricts by default

Zorka.Agency · Product Lead · March – November 2025

FemFast is a personalized AI-driven wellness app for women that aligns fasting, workouts, and supplements with hormonal cycles. The task: ship an MVP that would justify further investment in 60 days. No team expansion. In a category Meta restricts for advertising by default.

I ran 40+ JTBD interviews with the target segment before writing a single requirement. That produced sharp segmentation, and segmentation trimmed the MVP to what the smallest coherent audience would actually pay for. The alternative was obvious — ship the founder's original wish-list at half quality.

Then, measurability. A paywall analytics stack on RevenueCat, Mixpanel, and AppsFlyer. SKAdNetwork and Meta's AEM to measure conversions on iOS 14+ without breaking privacy defaults. And a sustained approval process with Meta for advertising in a restricted category — access at the category level. Not a one-off clearance for a single ad.

The tally

MVP shipped in 2 months, on time, no team expansion · onboarding completion 33% → 75% · CAC −16% · organic installs 7.9% → 23.7% · SKAdNetwork + AEM unlocked 78% of previously unavailable events · ×3 UA volume at stable CPI

Both cases ran under someone else's brand, so you'd still have to take my word for them. The next exhibit — you wouldn't.

Beat 04 Exhibit C

What you can download right now

A product manager who has never shipped anything of their own doesn't know the price of their own decisions.

I ship products under my own name, not just under a job title. Four live, several in the pipeline. The links below lead to the stores — where I control nothing.

HandoverBoat

handoverboat.com · charter deposit protection · flagship

A charter deposit protection platform for yacht and boat rentals. Before you sail: photograph the boat's condition with GPS-timestamped, RFC 3161–qualified evidence. When you return: a matching inspection produces a before/after PDF the operator and the charterer both sign against. A public verify endpoint lets anyone independently confirm the record wasn't edited.

Built solo, top to bottom: system architecture, Next.js 15 + Supabase + Stripe stack, UI/UX, illustrations, product design, legal review across HR/EL/IT/PL jurisdictions, copy, SEO, launch. Eight languages live at launch.

Positioning: an independent evidence record with a reversal of the burden of proof — not a "you-signed-so-you-accepted" acceptance engine.

Sixteen zones on one boat illustration, each a photo-capture point.

Three steps to the first skewer, in RU / EN / PL.

Shashlyk Assistant

mangal-time.shop · consumer app, multilingual, live in production

A shashlyk grilling timer and assistant. The interesting product decision: the UX is designed around the operator's real conditions — dirty hands, bright sun, awkward posture, split attention. Every touch target, every timer trigger, every glanceable state is tuned for that context. A calm office demo would have called for a different interface entirely.

The domain runs deeper than a timer. Meat type and cut, coal stage, open-flame behavior, fuel type, height above coals, skewer count — a grillmaster's tacit knowledge translated into structured inputs. Multilingual from launch across RU / EN / PL, illustrated in-house.

Built for friends; open to the world.

Tapset

tapset.fit · one-tap strength training log · phone + Wear OS

A strength-training log for people who already train and want to record a set faster than a spreadsheet allows. Weight and reps arrive pre-filled with the last result — match it, and logging is a single tap. The rest timer starts itself the moment a set is logged and vibrates through a locked screen in a gym bag. No account, no server: the entire history lives on the device, fully offline.

Tapset is also built for smartwatches running Wear OS. The watch takes over the between-sets loop — rep counter, one-tap confirm, rest countdown on the wrist — so the phone can stay in the bag from the first set to the last. The phone keeps the record; the watch is the remote.

Same design constraint as Shashlyk Assistant: an interface tuned for the actual moment of use — chalked hands, ninety seconds of rest, split attention.

Weight and reps arrive pre-filled from last time — a set logs in one tap.

ArtPaw

artpaw.shop · live e-commerce · real orders across the EU

Renaissance-style pet portraits — your dog or cat composed in the light of the old masters, printed, framed, and delivered across Europe. A full Shopify storefront: catalogue, variants, checkout, EU shipping and tax, and every policy page.

I built the whole thing — brand, copy, imagery, and store — 100% by AI-assisted coding, originally as an attribution testbed for my day job. Then it started taking real orders.

The point it proves: vibe-coding reaches all the way to a transacting business. It doesn't stop at a prototype.

The hook: your pet, painted like royalty, framed and shipped across the EU.

The products you can touch. The method you can't — so I put the method out in the open too.

Beat 05 Exhibit D

What you can install with one command

A method you can't hand over isn't a method — it's personal luck.

I ran discovery at RedTrack, then packaged my own pipeline into a toolkit any product manager can install in one command. Fifteen sub-agents across twelve steps take a raw brief all the way to a structured, bilingual Notion report. I authored the agent network and the orchestration; the framework underneath is Ivan Zamesin's Advanced JTBD.

15
sub-agents
12
steps
4
research phases
RU · EN
bilingual output
Phase 01

Structure the brief

Phase 02

Parallel research — market map, competitors, community signals

Phase 03

AJTBD canon validation gate

Phase 04

Segmentation, QA, and the bilingual report

See the discovery flow View the code on GitHub Install from Skills.sh

Install in one command · MIT-licensed · open source · links open in a new tab

Four exhibits. Now — why I called them exhibits at all.

Beat 06 The Turn

CVs get read.
Work gets verified.

I've been calling the four blocks above exhibits. That's not a stylistic tic.

Before product, I was a lawyer. A Master of Law from Belarusian State University, then two years at a bank — legal counsel, then senior legal counsel. Law runs on an order that is not up for debate: the evidence comes first, then the claim. Reversed, it works nowhere.

I carried that order into product wholesale. That's why 300+ customer interviews come before requirements. That's why the agents sit on schedules and write logs. That's why the apps live in the stores and the method lives in an open repository. That's why HandoverBoat has a public verify endpoint: anyone outside can confirm a record wasn't edited — including someone who doesn't trust me personally.

The method has a name too: Advanced Jobs to be Done and Next Move Theory — Zamesin's canon. It runs on one principle: adoption follows evidence.

You don't need to trust me. Do you have three minutes?

Vlad Derkach

Five years in product. Twelve building businesses before and around it: law, sales, and companies of my own. CPO once. 300+ customer interviews.

I sail. Not as a hobby garnish — sailing teaches the same things a product organization needs at scale: teamwork, resilience, focus, and calm under pressure.

Beat 07 Verification

The three-minute verification

Nothing below requires my participation — or my honesty.

  1. 01
    Install my Discovery FlowOne command. Fifteen agents. MIT-licensed — read the code before you run it.
    GitHub ↗ Skills.sh ↗
  2. 02
    Download HandoverBoatCharter deposit protection for yacht rentals. Eight languages, works offline.
    App Store ↗ Google Play ↗
  3. 03
    Verify someone else's inspection recordA public verify endpoint: anyone can confirm a record wasn't edited — not just the parties to the deal.
    handoverboat.com ↗
  4. 04
    Download Shashlyk AssistantAn interface for dirty hands and bright sun. RU / EN / PL.
    App Store ↗ Google Play ↗
  5. 05
    Log a set in TapsetA one-tap strength log. Phone + Wear OS watch, fully offline.
    App Store ↗ Google Play ↗
  6. 06
    Reach checkout on ArtPawA live store with real EU orders. Built 100% with AI-assisted coding.
    artpaw.shop ↗

If questions survive that, they'll be specific ones. Bring the specific ones.

Verification shows what I've done. The one thing it won't show is how I'd think in your context. That's the next beat.

Beat 08 Method

How I think

Three operating ideas in one chain: where the jobs come from, where the mechanics come from, how the value gets delivered.

Each one has a limit, and I've named it right underneath.

Idea 01

Where the jobs come from

You can't automate the interview. You can automate everything around it.

Nothing replaces a live customer interview — I tested that conclusion hard before accepting it. A synthetic persona generates what's plausible, and plausible is the opposite of what you go into discovery for. But most companies already sit on an archive of real conversations with real customers: sales calls, demos, onboarding, support. Those are authentic words from real people, collected for a different purpose. What's left for the model there is the extractor's role: the context, the trigger, the success criteria, the alternatives that were weighed — it has all been said already, and only needs pulling out.

Given the discipline to store those conversations properly and a lawful basis to use them, transcripts give up Jobs at scale. That produces a map instead of a handful of points — and live interviews stop being spent on reconnaissance.

Where it stops. It will not surface hidden, unarticulated jobs — those still come out live. The archive is also biased towards people who made it to a conversation at all. And with no lawful basis to use the recordings, the method doesn't run.

Idea 02

Where the mechanics come from

Shop for mechanics on a donor market.

Benchmarking direct competitors converges a whole category on the same dashboard. Everyone studies everyone, so nobody finds a mechanic that isn't already in the room. That's a local optimum, reached expensively.

So I look for a more mature market whose users behave the same way, and go shopping there instead. A performance media buyer works several screens with their own money at risk in real time, decides within the day, and distrusts somebody else's automation. So does a day trader — and trading terminals are decades of product selection already paid for by a bigger, older, wealthier market. Thresholds instead of reports. Watchlists. Acting straight from the alert rather than being sent to an interface. Real-time P&L. Stop-loss limits. Deliberate information density where the design instinct says to simplify.

Where it stops. An analogy is not an identity — decision horizons, switching costs and regulation all differ, and part of the segment genuinely works slowly and in batches. Every transfer is a value hypothesis, and it goes through the same cheapest-test-first validation as anything else.

Idea 03

How the value gets delivered

Silence is a feature.

There have been two eras of value delivery. In the first, the user had to come in and do the work. In the second, products started doing the work and delivering a result — reports, digests, recommendations. Both still charge the user attention: somebody has to read the thing and decide whether it mattered.

The third shift is about less attention. The product stays quiet while everything is fine, and speaks only when an action is required — carrying the action itself, along with the news. A recurring digest is actively harmful here: if nine messages in ten need nothing, it trains people to ignore the tenth. Zero false alarms beats completeness. The metric is the share of messages followed by an action; open rate doesn't count.

Where it stops. Staying quiet means occasionally missing a weak signal. I take that trade knowingly: a missed alert costs one mistake, while trained-in indifference costs the entire channel — and that one is irreversible.

All three rest on the same base — Advanced Jobs-to-be-Done, in Ivan Zamesin's canon rather than the generic Christensen framing: a segment defined by shared jobs and success criteria, value as energy saved rather than features added, and the riskiest assumption tested before anything gets built.

That's my own account of how I work. Next — the versions from people who watched it from the outside.

Beat 09 Witnesses

What the people who worked with me say

"The way he frames problems now — that's strategic-level thinking."
Vlad Zhovtenko, CEO, RedTrack.io
"He is the kind of person who turns complex and risky startup hypotheses into functioning businesses."
Dzmitry Khudy, CEO, Zorka.Agency
"When budget cuts hit Roketo, Vlad secured grants, retained the team, and led our pivot as CPO. A highly reliable product leader."
Taras Dovgal, Co-Founder at NoVPS · previously at Roketo

Speaking

  • Product Summit, Sopot
    Product Discovery automation with AI agents. Seminar participation, in Polish.
    June 2026
  • Internal training, Warsaw
    Using code to automate product workflows and build autonomous / human-in-the-loop agents.
    May 2026
  • E-commerce Expo, Berlin
    On-site customer discovery for MarTech.
    February 2026

Beat 10 Caveats

Caveats

A page where everything adds up is a bad page. Here's what doesn't add up on this one.

  • Part of the work sits under NDA. A year at a crypto neobank across the EU and LATAM I can't describe. It's on the track record; the details won't be.

  • I don't publish internal operating numbers. Exactly how many tickets the Support AI Assistant has processed is RedTrack's data — mine to use, not to disclose. I'll name the order of magnitude; the exact figure I won't.

  • I take one or two engagements at a time, alongside my main role. The fractional format runs three months minimum — shorter pays off for nobody.

  • Discovery isn't sold separately. Research, segmentation, and opportunity mapping ship inside every engagement rather than as a standalone service.

  • My method is opinionated. Zamesin's Advanced JTBD rather than the generic Christensen framing. If your company runs a different canon, we will argue — and I will argue back.

Beat 11 Work

How to work with me

The work is always the same shape: production AI inside a product organisation — agents that keep running after I leave.

Two weeks

AI Product Audit

I map where AI actually pays inside your product and operations, which agents to build first, and what each will cost per user per month. You get a prioritised build list with unit economics attached — not a maturity model.

From €4,000
Four to eight weeks

Agent Fleet Build

I design and ship the first agents into your production operations — support, competitive intelligence, research, onboarding — with logging, evals, confidence thresholds and a cost model from day one. Your team takes over a running system and a handover document.

From €12,000
One to one and a half days a week

Fractional AI Product Lead

I own the AI product line: what gets built, what gets killed, how quality is measured, and how the team learns to work this way. Minimum three months.

From €5,000 / month

The full track

Eleven companies, four industries — five years of it as a product manager. Details on LinkedIn.

  • RedTrack.io
    Product Discovery Lead & AI Champion
    AI product operations · 25+ agents in production
    Nov 2025 – Sep 2026
  • Zorka.Agency (FemFast)
    Product Lead
    Mar 2025 – Nov 2025
  • ERA Wallet (ex-1inch)
    Senior Product Manager
    Sep 2024 – Mar 2025
  • Crypto NeoBank (NDA)
    Product Lead
    Aug 2023 – Aug 2024
  • Roketo
    Chief Product Officer → Product Manager
    May 2022 – Aug 2023
  • NOWPayments
    Product Manager
    May 2021 – May 2022
  • ChangeNOW.io
    Product Consultant
    May 2021 – Sep 2021
  • Belkasoft
    BizDev / Product Development Manager
    Oct 2017 – Mar 2021
  • Smart Policy
    Business Development Manager
    Nov 2016 – Oct 2017
  • Labs Publicity Group
    Production Manager
    Jul 2014 – Nov 2016
  • Bank «Reshenie»
    Senior Legal Counsel → Legal Counsel
    Aug 2012 – Jul 2014

Beat 12 Contact

You've read the fortieth CV this week.
Now verify one.

Open to: Head of AI Product, AI Product Lead, and Principal PM (AI / agentic products) roles — B2B SaaS, scale-ups and Series A/B. Remote or Gdańsk / Tricity.

Name: Vlad Derkach in everyday use, Uladzislau Dziarkach on my Belarusian passport and on my App Store listings — the same person.

Based in Europe · remote-first.