Artyom Zankevich
Portrait of Artyom Zankevich

About Artyom Zankevich

I'm an AI engineer and engineering leader at Akveo in Vilnius. I prototype AI agents myself, then lead the 30+ engineers who ship them. In my department, commits written with AI tools rose from 6% to 74%.

Engineering
AI Engineer, Akveo
Leadership
Head of Web Development, 30+ engineers
Focus
AI agents, voice agents and AI-assisted development
Stack
TypeScript, OpenAI, Anthropic Claude, MCP
Based in
Vilnius, Lithuania, remote
Writing code since
2014
Education
Computer Science, Belarusian State University
Languages
Belarusian (native), Russian (native), English (professional)

AI in Engineering

Since 2024 I have been moving the department to AI-assisted development, with the same review and tests for code written with AI tools. The line is the share of commits written with AI tools; the numbered marks are the steps that changed how the department works.

Commits Written with AI Tools
Q1 ’24: 6%, Q2 ’24: 8%, Q3 ’24: 10%, Q4 ’24: 13%, Q1 ’25: 16%, Q2 ’25: 20%, Q3 ’25: 27%, Q4 ’25: 38%, Q1 ’26: 52%, Q2 ’26: 64%, Q3 ’26: 74%. Numbered marks are listed below the chart.0%25%50%75%100%
  1. 1AI Adoption ProgramMar 2024
  2. 2Engineering Workflow PluginsJun 2025
  3. 3Voice Support AgentAug 2026

Commits that name an AI co-author are only a floor, since tools like Cursor and Copilot leave no trace.

Projects

AI agents I built for pre-sales, the tools my engineers use every day and the products I built or led, each with what it changed.

Selected work first, then the rest by date. I keep client names private and describe each project by what it was and who used it. Open a row for the details.

Show
Order Confirmation AgentAI prototype for a distributor's purchasing teamClientAIAI engineerAkveo, 2026A working prototype in 3 days, where a proof of concept used to take 4–6 weeks
Problem
Buyers checked supplier order confirmations against purchase orders by hand, line by line.
Team
Solo, with a pre-sales lead
My part
Discovery, prototype, demo
What I did
  • Extracted order lines from supplier emails and PDFs with structured outputs
  • Matched them to purchase orders with RAG over the product catalog, and flagged every mismatch
  • Prepared the matched orders for SAP import, with a review screen for the buyer
Stack
Anthropic ClaudeStructured outputsRAGTypeScriptNestJSPostgreSQLSupabase
Voice Support AgentAI prototype for an electric vehicle charging network's driver supportClientAIAI engineerAkveo, 2026Demoed to the prospect as a working product before the contract
Problem
Drivers call support when a charger fails, in many languages and at any hour.
Team
Solo, with a pre-sales lead
My part
Discovery, prototype, demo
What I did
  • Built an agent that answers drivers by voice in their own language, on the OpenAI Realtime API
  • Gave it tools through MCP to check a station, restart a charging session and open a ticket
  • Handed the call to a person, with a summary, when the agent could not help
Stack
OpenAIReal-time voiceTool callingMCPTypeScript
Proposal PluginInternal pre-sales tool on Claude Code and CodexInternal toolAIAI engineerAkveo, 2026 – nowBrief to first draft in 1 day instead of 3+; used on 6 pre-sales deals in the first three months
Problem
Every proposal started from a blank page and took days of back and forth.
Team
Solo, with pre-sales leads
My part
Design, build, rollout
What I did
  • Designed skills that shape a solution with the proposal author, then build the proposal from it
  • Pulled sources from the CRM and team chats through MCP, and similar cases from the department's library with RAG
  • Rendered a reviewed HTML proposal with an optional PDF on the approved templates
Stack
Claude CodeCodexMCPRAGPrompt engineering
Engineering Workflow PluginsPlugins and skills for Claude Code and CodexInternal toolAILead engineerAkveo, 2025 – nowPart of every team's daily workflow
Problem
Teams repeated the same setup, review and documentation steps by hand on every project.
Team
Used across the department
My part
Design, build, rollout
What I did
  • Built Claude Code and Codex plugins and skills for project setup, code review and documentation
  • Added an MCP server for YouTrack, so agents read and update the tasks they work on
  • Kept each plugin to one job, with a usage guide
Stack
Claude CodeCodexMCPTypeScript
AI Adoption ProgramA change in how the whole department buildsInitiativeProgram leadAkveo, 2024 – nowCommits written with AI tools went from 6% to 74% across a 30+ engineer department, 2024 to 2026
Problem
Make AI-assisted development a normal, reviewed part of how every team works.
Who
30+ engineers
My part
Strategy, rollout, measurement
What I did
  • Picked the tools and rolled them out team by team
  • Wrote guidelines on what to use AI for and how to review its output
  • Measured the effect on speed, quality and cost
Areas
AI adoptionClaude CodeCodex
AI Interview CoachiPhone app for interview practiceSide projectAIAI engineerPersonal, 2026 – nowIn TestFlight beta
Problem
Practice a job interview for a specific role and get honest, consistent feedback on every answer.
Team
Solo
My part
Product, iOS, backend, AI
What I did
  • Built the voice interviewer on the OpenAI Realtime API
  • Scored answers against versioned rubrics with structured outputs, and checked every prompt change with evals
  • Wrote the native SwiftUI app and its NestJS and PostgreSQL backend
Stack
OpenAIReal-time voiceStructured outputsEvalsSwiftSwiftUINestJSPostgreSQLRailway
Language Tutor PluginOpen-source plugin for Claude Code and CodexSide projectAIEngineerPersonal, 2026 – nowOpen source on GitHub
Problem
Turn my own textbooks and audio into lessons an AI tutor can prepare and review.
Team
Solo
My part
Design and build
What I did
  • Built skills that prepare each lesson from local textbooks and audio and keep progress between sessions
  • Added a short recap after each lesson for later review
Stack
Claude CodeCodexPrompt engineering
AI Meetups at AkveoWeekly talks and hands-on sessions for Akveo engineersInitiativeCo-organizer and speakerAkveo, 2026 – now12+ sessions
Problem
Engineers needed a place to learn AI-assisted development from each other's projects.
Who
Open to all engineers
My part
Program, talks, demos
What I did
  • Co-organized the program and gave talks and demos
  • Showed cases from the department's own projects instead of vendor demos
  • Kept notes and recordings in the knowledge base
Areas
AI adoptionClaude CodeCodex
Internal ToolsSmall apps for Akveo's own operationsInternal toolEngineerAkveo, 2025 – nowAbout 20 hours of manual work saved a month
Problem
Routine work in hiring, time tracking and billing took hours of manual steps every month.
Team
Solo
My part
Design, build, rollout
What I did
  • Built a time log calendar widget for YouTrack and an app for monthly invoices and time reports
  • Synced job openings from the hiring system to the careers page
Stack
TypeScriptNext.jsReactCloudflareClaude Code
Hiring and Onboarding PracticeHow the department hires and onboardsInitiativeHiring leadAkveo, 2021 – now75+ engineers brought in
Problem
Hire and onboard fast without lowering the bar while the department doubled.
Who
Hiring panel
My part
Assessment, onboarding, allocation
What I did
  • Set a technical assessment with a clear bar, now with AI-aware tasks
  • Wrote an onboarding plan for the first weeks
  • Matched engineers to projects, cutting the time to staff one from 21 to 7 days
Areas
HiringEngineering management
Earlier projects, 2014 – 20217Show
Emergency-Travel MembershipHealthcare app for web, iOS and AndroidClientTeam leadAkveo, 2019 – 2021
Problem
Members of an international medical-transport program manage plans, dependents and billing from anywhere.
Team
4–6 engineers
My part
Frontend and mobile lead
What I did
  • Led the frontend and mobile team, and built the member portal and the mobile app
  • Turned client requests into specifications the team could estimate
  • Reviewed the user experience and improved it
Stack
AngularIonicCapacitorNgRxNebularAWS
Machine Learning Workflow PlatformWeb platform for machine-learning teamsClientSenior engineerAkveo, 2019
Problem
A Silicon Valley startup's platform for machine-learning and deep-learning workflows needed a fast, server-rendered interface.
Team
3–4 engineers
My part
Frontend
What I did
  • Built the frontend and improved the user experience
  • Added server-side rendering with Angular Universal
Stack
AngularAngular UniversalAngular MaterialNebular
Home-Services MarketplaceConsumer marketplace on the webClientSenior engineerVention, 2018 – 2019
Problem
Match thousands of customers a week with vetted cleaners and handymen through a 60-second booking flow.
Team
Large product team
My part
Fullstack
What I did
  • Built frontend and backend features
  • Implemented the interface from the designs
  • Improved SEO on the public pages
Stack
Ruby on RailsReactJavaScript
In-Store Rewards and ReferralsRewards web app for small shopsClientEngineerVention, 2017
Problem
Give small shops a digital rewards and referral program with automated marketing.
Team
2–3 engineers
My part
Frontend
What I did
  • Upgraded Ember.js and the authentication flow
  • Built frontend features
Stack
Ember.jsJavaScript
Web Trading PlatformBrowser trading platform for active tradersClientSenior engineerVention, 2016 – 2017
Problem
Active traders need real-time markets, charting, trading and news in the browser without lag.
Team
Frontend team
My part
Frontend, architecture
What I did
  • Estimated the work and reported on progress across the full cycle
  • Changed the architecture for real-time data
  • Refactored the app and made it faster
Stack
AngularTypeScriptWebSocketsKarma
Talent Booking and SchedulingBooking app for the talent industry, on iOS, Android and webClientTeam leadVention, 2015 – 2016
Problem
Agencies and models schedule appointments and events from one app on phones and the web.
Team
4–5 engineers
My part
Frontend and releases
What I did
  • Led the full development cycle, with reports and estimates
  • Released to the App Store and Google Play
  • Reviewed code and ran maintenance and support
Stack
Ruby on RailsAngularIonicCordovaRabbitMQ
Research Data CrawlerData pipeline for a research-information productClientEngineer1science, 2014 – 2015
Problem
Extract research texts from web sources automatically and at scale.
Team
Solo, with a data team
My part
Crawlers
What I did
  • Designed and maintained Scrapy crawlers
  • Ran distributed extraction on Scrapinghub
  • Made the crawlers faster and easier to scale
Stack
PythonScrapyXPath

Engineering Leadership

I run web engineering at Akveo, its largest production department. I hire engineers, decide who works on which project and make sure the work ships. Since I took it over in 2021 the department has worked on 70+ client projects, 15 to 19 of them at any time.

Results

People

  1. Engineers in the departmentSince Aug 2021, with the same hiring bar: from 15 to 30+
    ×2
  2. Average time in the departmentJan 2024 to Oct 2026: from 1.8 years to 3.4 years
    +89%
  3. Engineers brought in a yearIn-house and partner engineers, 75+ since 2021: from 13 to 19
    +46%

Delivery

  1. Bench timeShare of engineer time between projects, 2023 to 2026: from 9% to 4%
    −5 pp
  2. Billable utilizationShare of engineer time on client work, 2021 to 2026: from 73% to 92%
    +19 pp
  3. Time to staff a new projectFrom signed contract to a team on the project: from 21 days to 7 days
    −67%

AI

  1. Commits written with AI toolsShare of all commits, 2024 to 2026: from 6% to 74%
    +68 pp
  2. Time to a working AI prototypeA pre-sales proof of concept, built with AI agents: from 4–6 weeks to 3 days
    −91%
  3. Client proposal, from brief to draftWith the proposal plugin: from 3+ days to 1 day
    −67%
How It Runs
1. Hire: Technical assessment and a clear bar (AI-aware tasks) 2. Onboard: First project ready in weeks (AI tools from day one) 3. Match: Skills, growth goals and business needs (AI skills in the profile) 4. Deliver: Oversight and unblocking teams (AI-assisted reviews) 5. Grow: Shared practices and internal talks (Weekly AI meetups)

How I Work

The rules I work by when I build with AI and lead teams, each with what it looks like in practice.

  1. 01

    Small Loops

    Build a prototype in days, then decide based on what it shows.

    Before a big commitment, I try the risky part first.

    How I Built My Personal Site with AI Agents
  2. 02AI

    Evaluate, Then Ship

    An LLM feature is ready when its evaluation says so, not when the demo works.

    Before launch, an AI feature has test cases with known answers, a cost per call and a fallback.

  3. 03AI

    People Own the Result

    AI can write the code. A person still answers for it.

    Every change goes through the same review and tests, whoever or whatever wrote it.

  4. 04AI

    AI Where It Pays

    Start where the gain can be measured, check it, and only then scale.

    Each new AI tool starts with a baseline and one number it has to beat.

  5. 05

    Estimate Honestly

    Show the range, the assumptions and the risks from the first day.

    Every estimate I send has a range and its three biggest risks.

  6. 06

    Teams before Plans

    A strong team with clear goals beats a clever plan.

    I match engineers to projects by their skills and what they want to learn, not just by who is free.

Experience

My engineering roles since 2014, at 3 companies. A line that runs on in dashes is a role I still hold. Ticks mark the client projects listed in Projects.

  1. AI Engineer at AkveoAug 2025 – nowProjects: Order Confirmation Agent, Voice Support Agent.
  2. Head of Web Development at AkveoAug 2021 – now
  3. Software Engineer at AkveoMar 2019 – Jul 2021, senior and team leadProjects: Emergency-Travel Membership, Machine Learning Workflow Platform.
  4. Software Engineer at VentionMay 2015 – Mar 2019, senior and team leadProjects: Home-Services Marketplace, In-Store Rewards and Referrals, Web Trading Platform, Talent Booking and Scheduling.
  5. Software Engineer at 1scienceDec 2014 – May 2015, contractProjects: Research Data Crawler.

Contributions and Training

  • FlexGet

    Content automation tool

    My open-source contributions include FlexGet, a Python tool for collecting and downloading content such as podcasts, comics and series. It connects RSS feeds, websites and other sources through plugins, with configurable rules that filter content and pass it to download tools.

    FlexGet on GitHubOpen source

  • Management Skills Development Program

    Stratoplan Management School

    The program covered the manager's role during growth and development crises, motivation and feedback, and constructive confrontation in difficult conversations with employees. It combined theory with practical work.

    View Stratoplan certificate (PDF)Training

  • Nebular

    Angular UI library

    I worked on Nebular, an Angular UI library for building applications with reusable components and a consistent design. Based on the Eva Design System, it includes customizable light and dark themes, along with authentication and security modules.

    Nebular on GitHubOpen source

  • CVAT

    Computer Vision Annotation Tool

    I contributed to CVAT, an open-source platform for preparing image and video datasets for computer vision. It brings annotation tools, automated labeling and team review workflows together to help turn raw data into training datasets.

    CVAT on GitHubCVAT on NASA ADSOpen source

Now

What I am working on at the moment.

  • Building. Working AI prototypes for pre-sales, and an AI interview coach for iPhone on the side.
  • Running. Weekly AI meetups at Akveo, which I co-organize; 12+ sessions so far.
  • Writing. Building my site with AI agents, and how I estimate software projects.How I built this site

Updated Oct 2026

Contact

Reach me on LinkedIn or X, or download the CV.

You can also find my answers on Stack Overflow.

Download CV PDF