• Файл

Pavlo

Java-розробник

Вік:
27 років
Місто:
Київ

Контактна інформація

Шукач вказав: Телефон

Прізвище, контакти та світлина доступні тільки для зареєстрованих роботодавців. Щоб отримати доступ до особистих даних кандидатів, увійдіть як роботодавець або зареєструйтеся.

Завантажений файл

Версія для швидкого перегляду

Це резюме розміщено у вигляді файлу. Ця версія для швидкого перегляду може бути гіршою за оригінал резюме.

Pavlo Serikov [відкрити контакти](див. вище в блоці «контактна інформація») · [відкрити контакти](див. вище в блоці «контактна інформація»)
Kyiv, Ukraine (remote) · GitHub
Java Engineer · Agentic AI Systems

> PROFILE

I'm a Java engineer at heart with hands-on experience building and running agentic AI systems,
nearly 3 years of hands-on Java development (since 2024) and 6 years of taking projects from
requirements to delivery. I came to software from architecture, and I still think like an
architect: structure first, the people who will use what I build, and what it costs to build and
run. On my current project I grew from Java developer into the maintainer of an agentic AI
platform and took it with the team from prototype to regular production use: 8 cooperating LLM
agents with an orchestrator, shared memory, RAG, MCP tools and human approval for any action
that changes real data. I own the repository, the AWS deployment, auth and agent security, and I
built a calculation agent together with engineers from Microsoft. I've worked with Claude,
OpenAI and Gemini models, review code, mentor a junior engineer and present architecture to
stakeholders. English C1.

> SKILLS

Agentic systems Multi-agent orchestration (router + specialist subagents), tool calling, state
and long-term memory, step/recursion limits, human-in-the-loop approval,
handoff to humans
Frameworks & MCP (FastMCP servers, Java MCP tools), LangGraph, LangChain (AWS Bedrock),
protocols LangSmith feedback
Models Anthropic Claude (via AWS Bedrock, SigV4 LLM proxy), OpenAI GPT, Google
Gemini; model and endpoint configuration per agent, token and output limits
RAG & knowledge RAG over product documentation and a Bedrock Knowledge Base, Gemini-based RAG,
per-user access to knowledge via OAuth
AI safety & control Least-privilege agent tools (read-only by default, mutations behind approval),
dry-run validation, output guardrails, per-user identity and tenant isolation,
SSRF allowlisting
Backend Java 21 (main language), Spring Boot, REST APIs, JSON/XML, PostgreSQL, SQL,
Kafka; Python (FastAPI, FastMCP, LangGraph) with AI coding tools
Platform & ops AWS (EKS, Bedrock, Aurora PostgreSQL), Kubernetes, Helm (dev/qa/prod), Istio,
Docker, GitLab CI/CD, diagnostics and log-based incident investigation
Security Keycloak, OAuth 2.0 (client credentials), OIDC + PKCE, signed-JWT verification
Leadership Architecture reviews, task decomposition, code review, branching and CI
standards, mentoring

> EXPERIENCE

Java Developer → AI Agentic Platform Maintainer · Luxoft 2025 – Present · Remote
Client: Halliburton Landmark DecisionSpace (DecisionSpace 365 / Geosciences) — "Harmony", a self-
hosted multi-agent AI platform inside an enterprise desktop application.
Prototype to production: own the platform's GitLab repository and pipelines; took it with the team
from demos to regular production use, and keep developing it.
Multi-agent system of 8 agents: an orchestrator classifies intent and routes to specialists for
documentation Q&A, project state, display control, geoscience calculations, external compute jobs,
petroleum-engineering knowledge (RAG) and geomodeling advice; agents are defined declaratively and
share long-term memory.

Calculator agent, built with Microsoft engineers: Java tools call the calculation engine's HTTP
API; every plan is dry-run validated (bindings, compatibility, function catalog, schema), and
execution waits for human approval.
Control of agent actions: read-only tools run unattended, while every tool that changes live user
data sits behind human approval (calculator, display and external-compute agents); output-length
guardrails and per-agent step limits.
Security and data isolation: closed a multi-user data-privacy gap end to end: desktop OIDC/PKCE
login, signed-JWT verification and OAuth 2.0 client credentials between agents' MCP servers,
tenant resolution from per-user JWT claims, Istio JWT policy and SSRF allowlist.
Integrations through MCP: a self-hosted connector exposes Java tools from the live desktop
session; standalone HTTP MCP servers serve documentation search, a knowledge base and external
compute; scaffolded the external-compute connector.
Platform and operations: self-hosted on AWS EKS (Helm dev/qa/prod, Istio, Aurora PostgreSQL,
Claude on Bedrock behind a SigV4 proxy); GitLab CI/CD with quality gates; per-service diagnostic
and restart jobs; log-based investigation of failures; end-to-end platform upgrades.
Feedback loop: thumbs-up/down feedback to LangSmith and stable live progress and approvals in the
LangGraph service (Python and TypeScript work delivered with AI coding tools, owning design,
review and testing).
Technical leadership: presented the release architecture review to client stakeholders, set
branching and CI conventions, review merge requests and mentor a trainee; work with the
architecture team and Microsoft engineers.

Architect & Interior Designer · Klinker Terra Design 2020 – 2023
Led projects from client requirements to delivery within budget; managed suppliers, including
German manufacturers, balancing quality and cost.

> PROJECTS

Lumina · Solo founder · iOS app, live on the App Store 2026
Screen-time app where children unlock apps by praying. Built and run entirely on my own: product and
UI/UX design, native iOS app (Swift/SwiftUI with AI coding tools), website, RevenueCat subscriptions
and pricing, Xcode Cloud workflows, App Store release, marketing.
Swift SwiftUI Screen Time API (FamilyControls) RevenueCat Xcode Cloud UI/UX design Website Marketing

EasyStay — Accommodation Booking API 2024
Team project: booking and property management with Stripe payments, Telegram notifications and
advanced search.
Spring Boot Spring Security JPA / Hibernate MySQL Liquibase Docker JUnit / Mockito

Online Book Store API 2024
Catalog, cart and order history with JWT authentication, SOLID layering.
Spring Boot Spring Security + JWT Spring Data JPA MapStruct MySQL Docker

> EDUCATION > COURSES & LANGUAGES

M.Sc. Urbanism & City Planning — Kyiv National 2023 – 2024 Mate Academy — Full-time Java
University of Building and Architecture Developer

B.Arch. Architecture — Kyiv National University 2016 – 2020 Sigma Software University — Java
of Building and Architecture Basics & Pro

English — C1

Схожі кандидати

Усі схожі кандидати

Кандидати у категорії

Кандидати за містами


Порівняйте свої вимоги та зарплату з вакансіями інших підприємств: