I build interfaces that stay instant while the model is still typing.

Senior Frontend & Agentic UI Engineer (Real-Time AI Systems)

Nine years in web development, the last three building conversational UI for AI developer tools over live LLM streams, the agentic tool-call workflows behind it, and the testing infrastructure that keeps them deployable. Deep TypeScript and modern Angular, plus production React.

Get in touch LinkedIn
Kraków, Poland Remote / hybrid · B2B contract · EU work authorization English · Ukrainian
  • 9 yrs
    Shipping production web software
  • 3 yrs
    Streaming UI & agentic workflows for AI tools
  • 1,600+
    Vitest integration tests adopted by 3 teams
  • 285 PRs
    Delivered to production with zero incidents
About

I own the hard part of AI products.

I am a senior frontend engineer based in Kraków, Poland. For the past three years, I have architected AI developer tools under long-term enterprise contract, and I take on select B2B contracts for teams building AI-native products. My work sits where a conversational interface meets a live model: keeping the UI instant and accessible while tokens stream, tools fire mid-answer, and upstream model APIs keep changing.

I architected an enterprise AI assistant frontend from scratch and evolved it to Angular Signals and Zoneless change detection. Decoupling UI state from streaming data is what made conversation branching, inline prompt editing, and mid-generation cancellation reliable in production.

I also focus on the parts that protect team velocity: a search re-architecture that cut maximum latency from 2 to 3 s down to under 600 ms, an integration testing harness that three adjacent platform teams adopted as their standard, and screen-reader support engineered directly for token-by-token output.

  • Real-time is a design constraint, not a feature

    Token buffering, throttled rendering and incremental markdown parsing, so code blocks and tables appear as they arrive without layout thrash.

  • Tests are what make AI features deployable

    Realistic service fakes instead of brittle mocks, so continuous deployment survives frequent upstream LLM API changes.

  • Accessibility is part of the build

    Anchored auto-scroll, skeleton and partial-response states, and live-region announcements built in from the first commit rather than retrofitted.

The stack behind all of it

what AI engineering roles hire for

Core
TypeScript Angular Signals Zoneless RxJS React Next.js Node.js
AI & real-time core
SSE streaming Tool-call UI Model Context Protocol RAG Hybrid search JSON Patch (RFC 6902) OpenAI & Claude APIs Local LLMs (Ollama)
Testing
Vitest Jest Testing Library Service fakes Storybook
Craft
Design systems Tailwind CSS Radix UI WCAG / ARIA Monorepos CI/CD
Work in progress

What I am building right now

Client work under contract, plus the products I build in the open. Full case studies follow on sergij.dev as the site goes live.

Enterprise AI assistant

2023 — present · under contract

Live

The real-time frontend layer of an enterprise AI assistant used by engineers at a global client. Built from scratch for the initial launch, then evolved to Angular Signals and Zoneless change detection to support complex tool-calling workflows, branching chats, and mid-stream cancellation.

  • SSE pipeline with token buffering, throttled rendering, and smooth incremental markdown parsing
  • Automatic reconnection and mid-stream resume to protect in-flight responses on dropped connections
  • Single shared component core powering both the standalone web app and the embedded desktop client
AngularSignalsSSEVitest

RoleReady

2026 to present · personal product

Building

A privacy-first career operating system. Professional achievements live as structured C.A.R.V. records (Context, Action, Result, Value). Give it a job description via URL or PDF, and it produces a targeted, ATS-optimized resume grounded strictly in stored facts. Local-first execution keeps confidential employer data safe from cloud prompts.

  • Resume tailoring reduced from 2 to 4 hours of manual effort to a single review pass under 45 seconds
  • Strict disclosure tiers split confidential experience from public variants, preventing NDA leakage
  • Interactive edits powered by RFC-6902 JSON Patch operations for non-destructive, surgical updates
Angular 22BunElysiaOpenAI & Ollama

sergij.dev platform

2025 — present · in progress

Soon

My own site, built as an AI-native content platform rather than a template. An agent drafts the content and I publish it, and the line between the two is held by the content store's own permissions and by build gates, never by wording in a prompt.

  • A draft-only agent identity over the Model Context Protocol: writes filtered to status=draft, publishing out of its reach, confidential career data deny-by-default
  • One model call per article returns three publishable drafts, behind a provider port with Claude, Gemini and an offline adapter so CI runs the pipeline deterministically
  • Zoneless signal-first Angular on Analog.js, statically prerendered, with Lighthouse SEO and accessibility gates that block the deploy instead of warning about it
  • A JSON-LD identity graph, llms.txt and feeds generated from the prerendered output, so assistants can quote the work rather than only crawlers index it
Analog.jsDirectusMCPClaude API
Experience

Nine years, measured in outcomes

Every line below shipped, and moved a number the business cared about.

  1. Aug 2023 — present

    Senior Frontend & Agentic UI Engineer (Real-Time AI Systems)

    SL IT DEVSERVICE · Kraków · remote · B2B contract for a global enterprise client

    • Architected the assistant frontend from scratch, then re-architected it on Angular Signals and Zoneless change detection as product and team grew.
    • Owned the streaming layer end to end: SSE with token buffering, throttled rendering, sanitized incremental markdown, automatic reconnection and mid-stream resumption.
    • Designed the integration-testing architecture and grew it to 1,600+ tests across 180+ suites; three adjacent platform teams adopted it as their standard.
    • Re-architected platform search: redundant requests down 40%, re-renders down 60%, maximum latency from 2–3 s to under 600 ms.
    • Delivered 285 production pull requests with zero incidents.
    • Led an enterprise-wide i18n migration across 5,800+ lines in one milestone, with custom AI CLI tooling for string extraction.
    • Built a design-system pipeline compiling 2,400+ SVG icons into webfonts and type-safe SCSS in a single command.

  2. Apr 2022 — Oct 2023

    Frontend Developer, Angular & TypeScript

    SL IT DEVSERVICE · Kraków · remote · consumer platform with an audience in the millions

    • Built discovery feeds and community activity walls for millions of users, with personalized content clusters and touch navigation.
    • Owned feed performance: virtualized lists, lazy-loaded media, reworked RxJS streams, with measurably higher time-on-page and click-through.
    • Replaced manual database edits with visual moderation dashboards, cutting curation time by over 70%.
    • Built an internal CMS for platform-wide campaign events, turning a multi-day manual process into an afternoon of work for one person.

  3. Apr 2020 — Mar 2022

    Frontend / Shopify Engineer

    Custom 4 Web · Kharkiv, Ukraine

    • Built a private "Product Upsell" Shopify app in React, Next.js and GraphQL with dynamic in-cart upsells and live inventory checks.
    • Engineered a filtering engine and multi-step prescription order form for an eyewear brand, cutting cart abandonment on custom lenses.
    • Shipped a cross-border localized experience with automatic plug detection, lowering return rates and lifting international conversion.

  4. Jun 2018 — Jan 2020

    Frontend Developer, Angular

    TradeInfo · Kharkiv, Ukraine

    • Replaced an unstable trading engine and CRM: full frontend migration in one month with zero client downtime, manager workflow steps down ~40%.
    • Led a redesign and self-service client portal in 3 months, lifting campaign conversion ~20% and ad ROI ~25%.
    • Automated publishing with a dashboard aggregating 5+ news sources, cutting release time from hours to minutes.

Building something that streams, thinks, or scales?

I am currently delivering on a long-term enterprise contract, and I take on select B2B contracts and key engineering roles for teams building AI-native products. If your conversational interface is lagging, token streaming feels unstable, or upstream model APIs keep breaking your UI, let's talk.

Email me Connect on LinkedIn
me@sergij.dev +48 698 893 410 sergij.dev Kraków, Poland · remote across EU and US