Artem Odulov

AI Product Owner Senior/Lead Business & Systems Analyst

AI-augmented product owner, business & systems analyst. I turn ambiguous requirements, stakeholder interviews, and manual delivery into working agentic workflows and measurable outcomes. 15+ years across enterprise systems, microservice integrations, and process re-engineering, with a track record of rescuing troubled deliveries.

Vilnius, Lithuania (Permanent Resident)
Explore my projects Download CV Detailed experience

AI in practice

Spec-Driven AI SDLC

Designed AI delivery lifecycles around living specs that stay in sync with the code and docs, so coding agents build from agreed requirements instead of ad-hoc prompts.

Agent Skills & Context Engineering

Packaged domain knowledge, design systems and team conventions into reusable agent skills and MCP integrations, giving every coding agent the same reliable context.

AI-Augmented QA

Built QA around agent skills and Playwright agents that generate and maintain tests, cutting the bug reopen rate by more than 6x.

Discovery to Spec

Turned stakeholder interviews and legacy modules into structured, versioned specs with LLM pipelines, shortening requirements gathering and giving QA a testable baseline.

AI Delivery Economics

Treated AI tooling cost as a delivery metric and cut token spend by 90% by moving from pay-per-token API billing to subscriptions.

Experience

AI Product Owner

AccuSourceHR (hybrid)
— present
  • AI-Augmented QA: Took on the QA role and built an AI-augmented QA framework with a library of agent skills, cutting the bug reopen rate by more than 6x.
  • AI Cost Optimization: Identified that pay-per-token API billing was driving AI spend and moved the team to subscription plans, reducing AI token costs by 90%.
  • Living-Spec SDD: Introduced a living-spec approach to Spec-Driven Development within the new AI SDLC, keeping specifications in sync with the code as the single source of truth for engineers and agents.
  • Requirements Baseline: Recovered and documented requirements for existing modules to migrate them to the spec-driven process and give QA a reliable basis for test design.
  • Design System: Created the design system for a new product and built it into the AI SDLC, so agent-generated UI follows it consistently.

Lead Business/ System Analyst

TOT Systems (remotely)
Feb. 2020 — Dec. 2025 (5 y. 10 m.)
  • Architecture & Scalability: Architected 4 core platform modules and over 15 microservices, serving as the technical backbone for a dozen distinct client systems.
  • Process Re-engineering & AI Adoption: Pioneered cross-functional AI adoption into engineering workflows, QA automation, and documentation drafting, directly boosting overall development productivity.
  • Delivery & Crisis Management: Rescued and stabilized troubled project deliveries by facilitating stringent requirements elicitation and aligning complex stakeholder expectations.
  • Standardization: Developed proprietary design templates and established a systematic software design process to ensure clear functional specifications and predictable development outcomes.

Business Analyst

Sxope (US, remotely)
Sep. 2024 — Feb. 2025 (6 m., part time)
  • Product Roadmap: Defined and guided the technical feature roadmap for a critical Business Intelligence (BI) product used by Executive Officers.
  • System Design: Designed and oversaw the implementation of an MVP for organizational structure-based permission management.
  • Quality Assurance: Designed Benchmarks service, performing functional validation (UAT) to ensure delivery quality.

Chief Business Analyst

Naumen
Feb. 2016 — Feb. 2020 (4 y.)
  • End-to-End Ownership: Led analysis and requirements engineering for large-scale enterprise systems, prioritizing product backlogs and taking full accountability for product outcomes.
  • Process Engineering: Engineered dozens of complex business processes and successfully redesigned/ implemented several core systems from scratch.
  • Leadership: Progressed from Analyst to Chief Analyst. Mentored junior colleagues.
  • Research & Publications: Co-author of the article "Disambiguation Solution for Persons' Accounts in Research Information Management Systems".
Earlier experience 2010–2016
  • Megafon Retail — Senior Business Process Analyst Aug 2015 – Feb 2016

    Modeled ERP business processes (SAP, 1C, MS SharePoint) down to data and attributes from scratch in ARIS.

  • Workservice — Automation Project Manager Jul 2014 – Aug 2015

    Managed the nationwide rollout of a cloud system across 50 regions for ~900 users.

  • OCRV — System Engineer Feb 2011 – Jul 2012

    Configured SAP R3 systems.

  • X5 Group — Business Process Analyst Jul 2010 – Dec 2010

    Modeled business processes using ARIS methodology.

Personal projects

A curated collection of my software design, development, and AI engineering pet projects. Web applications are deployed and accessible in the Pet Projects Hub.

HabitLog

Facts and stats, no judgment. A minimalist tracker designed for speed, clarity, and zero friction.

ReactSQLiteDrizzleWeb PushWebSocketsMCP

Linguasitu

Master real-world communication. Context-aware language packs and low-stakes AI chat rehearsals.

ReactNode.jsPostgresRedisIndexedDBTTSWhisper

GymFlow

Universal workout tracker with AI-powered feedback and workout program creation.

ReactMediaPipeRechartsIAM SSO

Movision

On-device camera motion analysis providing live auditory posture corrections during calisthenics.

ReactMediaPipeCanvas APIWeb Audio
More projects 5 more applications

HoopSnap

Real-time computer vision basketball trainer with automatic jump detection, reaction-time analytics, and localized responsive Material 3 design.

ReactMediaPipeM3 UIi18nIndexedDBPlaywright

Unisono

Collaborative workspace to harmonize wishes and plan activities together across small to medium groups.

ReactNode.jsWebSocketsMaterial UI

Beat Machine

Collaborative real-time drum sequencer and bass synth utilizing low-latency web audio generation.

ReactNode.jsWeb AudioWebSockets

HoopChart

Hands-free basketball practice tracking with spoken shot logging, custom drills, court-zone stats, and an optional on-device camera that measures shooting form.

LitTypeScriptCapacitorKotlinVoskMediaPipeDexie

Shopper

A shared, simple shopping list with a reusable catalog, real-time collaboration, and groups arranged in shop order.

ReactTypeScriptSQLiteDrizzleWebSocketsMCP

Expertise

Core Domains

Systems AnalysisBusiness AnalysisRequirements EngineeringSystems ArchitectureMicroservicesAPI & IntegrationsProcess Re-engineeringProduct OwnershipRoadmap ManagementUATSDLCRoot Cause Analysis

AI & Agentic Workflows

AI Workflow DesignAgent OrchestrationMCPPlaywright AgentsRAGSpec-Driven DevelopmentContext EngineeringPrompt Engineering

Modelling & Data

BPMN 2.0eEPCARISDBMLER-modelingJSONXMLSQLData Visualization

Tools

Jira Confluence Figma Stitch OpenDesign Claude Design Miro Postman GitLab Gitea Docker VS Code Orca Claude Code Codex OpenCode Cursor Antigravity N8N

Education

Specialist, IT management

State University of Management

Sep 2007 — Jul 2012

Deutsches Sprachdiplom, German language

Secondary school with in-depth knowledge of the German language

Sep 1997 — Jul 2007

Goethe Gymnasium (Rosslau, Germany)

Exchange program

Apr 2006 — Jul 2006

Personal

Languages

  • English — C1 (Advanced)
  • German — B1 (Intermediate)
  • Russian — Native

Hobbies

Basketball, Calisthenics, AI-dev, Non-fiction (IT, business, biology, astrophysics), Travels, Writing.

HoopChart unscripted run screen with make, miss, and undo controls and a court spot map
LitTypeScriptCapacitorKotlinVoskMediaPipeDexie
Android app with native voice controls and on-device form analysis

HoopChart

Hands-free basketball practice tracking with spoken shot logging, custom drills, court-zone stats, and an optional on-device camera that measures shooting form.

Description

HoopChart keeps the phone in your pocket during solo shooting practice. Call each make or miss, undo the last shot by voice, and follow custom drills or a free practice session. Review accuracy by court zone, drill, and date range, with local data storage and export/import for keeping your practice record portable.

An optional Form camera adds the other half of the picture: set the phone on a tripod and it measures how you shoot, such as jump height, knee bend, release timing, and arm extension, then compares those numbers for your makes and misses. Everything is analysed on the phone, and no images or video are kept.

Target Audience

Basketball players practicing solo who want to measure shooting progress without stopping to reach for their phone.

Architectural Highlights

  • UI: Lit and TypeScript with Material 3 web components inside a Capacitor Android shell.
  • Native voice: A Kotlin foreground service handles speech recognition with Vosk, spoken guidance, and session continuity while the screen is locked.
  • Form analysis: CameraX feeds MediaPipe pose landmarks to a pure Kotlin module that segments each jump shot, checks its quality, and computes the metrics. Capture keeps running with the screen locked, a live skeleton and framing guide help set up the tripod, and a "check framing" voice command confirms the shooter is in view.
  • Storage & analytics: Dexie over IndexedDB stores drills, shots, and resumable session state; statistics break down accuracy by court zone and date range. Form data lives in a private on-device store, keeps a compact pose window instead of video so shots can be re-analysed when the algorithm improves, and travels with export/import.
Shopper in dark mode with a seeded grocery list, group and item icons, and quantities
ReactTypeScriptSQLiteDrizzleWebSocketsMCP
Launch App
Google sign-in; access limited to provisioned users

Shopper

A shared, simple shopping list with a reusable catalog, real-time collaboration, and groups arranged in shop order.

Description

Shopper is a self-hosted shopping list for small groups. A persistent catalog makes recurring items easy to find and re-add, while ordered groups keep the list aligned with the route through a shop. Members see changes in real time, can undo their own actions, and can queue supported actions during a connection loss. An OAuth-protected MCP endpoint lets AI agents read and update authorized lists.

Target Audience

People sharing routine grocery lists who want fast list preparation and a clear, phone-friendly view in the shop.

Architectural Highlights

  • Frontend: React and TypeScript with TanStack Query, Material 3 styling, and an installable PWA.
  • Backend & storage: Express with Drizzle ORM and SQLite; Google sign-in and list membership control access.
  • Collaboration & agents: WebSocket updates and member presence, an offline action queue with conflict checks, and an OAuth-protected MCP server.
HabitLog Interface Screenshot
ReactSQLiteDrizzleWeb PushWebSockets
Launch App
Access restricted due to LLM costs

HabitLog

Facts and stats, no judgment. A minimalist tracker designed for speed, clarity, and zero friction.

Description

Unlike streak-obsessed trackers, HabitLog helps you objectively measure daily consistency without gamified pressure. It features goal-to-habit linking to compute an Impact Score,ing whether your habits actually improve your core goals (e.g. sleep vs energy). Pause habits via "The Bench" or archive them to keep data clean, request AI-powered statistics summaries with your own API key, and connect any MCP-compatible AI agent directly to your data via a built-in HTTP MCP server.

Target Audience

Data-driven individuals, professionals aiming for high performance, and anyone tired of standard habit tracking apps that trigger anxiety with lost "streaks."

Architectural Highlights

  • Frontend: React, TypeScript, and `@material/web` components for modern Material 3 styling.
  • Backend & DB: Express server with Drizzle ORM managing a local SQLite database.
  • Features: Web Push notification engine for daily logging reminders, WebSocket connection for real-time multi-device sync, CSV data mobility, AI-unraveling analysis integrations, and an HTTP MCP server so agents like Claude Desktop can log habits and check-ins autonomously.
Location: ../habitlog Command: npm run dev:all
Linguasitu Interface Screenshot
ReactNode.jsPostgresRedisIndexedDBTTSWhisper
Launch App
Access restricted due to LLM costs

Linguasitu

Master real-world communication. Any situation. Anywhere.

Description

Linguasitu bridges the gap between language study and reality. Users build tailored "Packs" of terms and phrases for exact situations (e.g. checking into an Airbnb, ordering at a pharmacy). High-quality Text-To-Speech (TTS) aids pronunciation, while live voice-first chat simulations with an AI roleplay partner build confidence before speaking in the real world.

Target Audience

Relocants integrating into new countries, travelers wanting to show respect by speaking target languages, and language learners seeking high-stakes context-driven study.

Architectural Highlights

  • Structure: React SPA monorepo frontend with a TypeScript Node.js backend.
  • AI & Speech Processing: OpenAI/Gemini situational vocab, local browser Whisper for Speech-To-Text (STT), and Edge TTS (Text-To-Speech) API.
  • Database, Cache & Storage: PostgreSQL database (Drizzle ORM), Redis TTS cache, and client-side IndexedDB persistence.
Location: ../linguasitu Command: npm run docker:up
GymFlow Interface Screenshot
ReactMediaPipeRechartsIAM SSO
Launch App
Access restricted due to LLM costs

GymFlow

Universal workout tracker with AI-powered feedback and workout program creation.

Description

GymFlow acts as a digital personal coach. It supports workout planning, logging history, and tracking body consistency metrics using Recharts. Additionally, it integrates a client-side computer vision engine that tracks user posture through the webcam and provides real-time rep counts and corrective form feedback.

Target Audience

Fitness enthusiasts, home workout practitioners, and powerlifters looking to optimize physical form, track volume history, and consult an AI training assistant.

Architectural Highlights

  • Frontend: React, TypeScript, Recharts graphing library, and Material Web Components.
  • Computer Vision: On-device MediaPipe Pose Landmarker for low-latency posture skeleton drawing.
  • Integrations: Integrated with a central IAM single sign-on auth service and supports i18next localizations.
Location: ../gymflow Command: npm run dev:all
Movision Interface Screenshot
ReactMediaPipeCanvas APIWeb Audio
Launch App

Movision

Advanced on-device motion analysis for perfect form.

Description

Movision is a zero-latency computer vision utility that assists calisthenics training. It tracks and analyzes body joints in real-time for 11 distinct exercises (including Push-ups, Squats, Dips, Planks, and Warrior II yoga). Users receive live range-of-motion metrics and immediate audio alarms/prompts when form slips, preventing injury.

Target Audience

Calisthenics athletes and gymnasts training at home who require immediate, hands-free feedback on skeletal posture and joint angles.

Architectural Highlights

  • Client Stack: React, TypeScript, and Vite.
  • AI Engine: Client-side MediaPipe Pose keypoint models overlaid onto custom Canvas coordinate systems.
  • Audio: Web Audio API oscillators producing low-latency sound cues for starting, counting, finishing, and form alerts.
Location: ../movision Command: npm run dev
HoopSnap drill setup with its basketball logo, orange styling, timer, and direction controls
ReactMediaPipeM3 UIi18nIndexedDBPlaywright
Launch App

HoopSnap

A real-time computer vision reaction and decision trainer, complete with skeletal tracking, automatic calibration, and deep analytics.

Description

HoopSnap is designed to run on a mobile device or tablet positioned on a basketball court. Utilizing on-device computer vision, it automatically calibrates the court floor plane, tracks player movement using a full pose-landmarker model with landmark smoothing, and detects vertical jumps. The app issues real-time audio/visual training cues and measures millisecond-level reaction times and rep counts. Results are synced to a local database and visualized via interactive progress charts, providing a closed feedback loop for reactive decision-making.

Target Audience

Basketball players, coaches, and sports trainers aiming to improve reaction speeds and decision making using consumer mobile devices.

Architectural Highlights

  • ML & Computer Vision: MediaPipe Pose Landmarker running on a worker thread, enhanced with isotropic distance mapping, dynamic floor plane calibration, and landmark EMA smoothing for jitter-free tracking.
  • Drill Engine & Analytics: Real-time reaction time tracking and rep counter, with post-drill session summaries, Dexie.js storage, and responsive Recharts dashboards.
  • Material 3 Design: Fully responsive navigation with a desktop side-sheet layout, custom M3 language picker, user preferences sheet (battery saver, debug overlay, guide sheet), and full i18next localization.
  • Quality Assurance: Complete test coverage featuring unit tests in Vitest and robust automated E2E suites via Playwright with mobile device emulation.
Location: ../hoopsnap Command: npm run dev
Unisono Interface Screenshot
ReactNode.jsWebSocketsMaterial UI
Launch App

Unisono

Real-time collaborative workspace to harmonize wishes and plan activities together.

Description

Unisono is a real-time collaborative workspace designed to harmonize wishes and opinions across small to medium groups of people. Users can spin up sessions, post virtual cards, sort ideas, vote on suggestions, and merge overlapping thoughts into unified plans. Perfect for groups of friends planning a trip, co-workers deciding on a team building event, or family members organizing activities.

Target Audience

Small to medium groups of people, such as friends planning something together, families organizing events, or teams looking to align their goals and reach quick consensus.

Architectural Highlights

  • Frontend: React, TypeScript, and Material UI (@mui/material) styling.
  • Backend & Realtime: Node.js Express server using WebSockets (`ws` package) for seamless multi-client cursor tracking and note synchronization.
Location: ../unisono Command: docker-compose up
Beat Machine Interface Screenshot
ReactNode.jsWeb AudioWebSockets
Launch App

Beat Machine

Expressive collaborative rhythm and sound synthesizer.

Description

Beat Machine allows multiple users to compose electronic music sequences together in real-time. It features a drum step grid (Kick, Snare, Hi-Hats, Ride) customizable from 4 to 64 steps, and a 2-octave polyphonic bass sequencer. Change tempo, toggle track mutes, adjust volume sliders, and see other users' cursors moving on the board in real-time.

Target Audience

Friends looking to collaborate on synth sequences, jam together, or draft rhythm sections online - just for fun.

Architectural Highlights

  • Frontend: React, TypeScript, and Tailwind CSS.
  • Audio Synthesis: Fully native sound synthesis using the Web Audio API (low-latency oscillators and panning nodes).
  • Backend & Realtime: Express server with a WebSocket (`ws`) server synchronizing playback, grid states, and mouse positions.
Location: ../ag-beats Command: npm run dev

Get in touch

Interested in working together or just want to say hi?

work.a.odulov@gmail.com Telegram LinkedIn