Roman Junior Java Developer
EN PL

Open to work

Backend developer building reliable services in Java.

I build REST APIs with Java and Spring Boot, cover them with tests, and take projects all the way to a running cloud deployment. Looking for my first role on a team where I can grow up.

Get in touch GitHub
Roman Tymanskyi
Status
Open to work
Focus
Backend · Java
Base
Opole, Poland
Stack
Java · Spring Framework · PostgreSQL· Redis· Docker
01

About

Junior Java developer focused on the backend. I got here through self-study and two side projects that I took all the way to working online services — JobTracker and Subtitle Translator — both from the data model to deployment.

What matters to me is not just assembling an app but being able to explain why each piece is there. I write tests (372 of them in JobTracker), work with Docker and managed cloud services, and read other people's code. Ready to relocate, work on-site or remotely.

Specialization
Backend / Java
Experience
Two side projects live
Languages
Polish C1, English B2
02

Projects

Full-stack · AI

A personal assistant for the job hunt: every vacancy and application in one place instead of a spreadsheet and a pile of bookmarks. You paste a job link and the system pulls the title, company, salary, location and requirements on its own; the vacancy lands on a kanban board you move from “Saved” to “Offer”. Built-in AI matches your skills against the posting, writes a cover letter, tailors your CV and prepares you for the interview.

Java 21Spring BootSpring AIPostgreSQL + pgvectorRedisReact 19TypeScript
Read more

Chrome extension · Backend

A Chrome extension that turns YouTube into a language trainer. Clicking any word in the subtitles gives a translation that accounts for the context of the line — no pausing, no dictionary in a second tab. Unfamiliar words go into a personal dictionary and come back for review on a spaced-repetition schedule; the extension also rates how well a video matches your current vocabulary.

Java 21Spring BootPostgreSQLTypeScriptChrome MV3Docker
Read more
03

Skills & stack

Languages

  • Java 21
  • TypeScript

Backend

  • Spring Boot
  • Spring AI
  • Spring Security
  • REST API
  • JPA / Hibernate

Data

  • PostgreSQL
  • pgvector
  • Redis
  • Schema migrations

Frontend

  • React 19
  • Vite
  • TanStack Query
  • Tailwind CSS

Testing & infra

  • JUnit
  • Testcontainers
  • Docker
  • Git
  • CI/CD
04

Contact

Open to offers and internships. Drop me a line — I reply fast.

Email rtymansktyi@gmail.com GitHub github.com/mnjkpf

Full-stack · AI

JobTracker

Java 21Spring BootSpring AIPostgreSQL + pgvectorRedisReact 19TypeScript

About the project

A side project I built entirely on my own — from the idea and the data model to a working online service. The goal was to make the job hunt organized and to automate the part of it that is normally done by hand.

How it works

You paste a link to a posting and the system collects the title, company, salary, location and requirements itself. The vacancy becomes a card on a kanban board whose stages you configure yourself.

  • Interesting → Applied → Recruiter replied → Interview → Offer
  • Old vacancies can be archived and brought back when needed
  • Statistics: applications sent, replies received, interviews held, offers landed

The AI assistant

AI here is not a chat window — it is a tool bound to one specific vacancy.

Skills analysis
Compares your CV against the posting: what matches, what is missing, and which requirements your experience already covers.
Cover letter
Generates a letter for that specific vacancy in a chosen tone. You can edit the text and ask for a new version that takes your edits into account.
Tailored CV
Produces a version of your CV for the posting, making relevant experience more prominent, scores it against ATS expectations, and exports to Word.
Interview prep
Generates likely questions from the posting and your background. Notes you write after an interview are stored and feed into later preparation.

Technical decisions

The key technologies were chosen for specific product problems, not out of habit.

PostgreSQL + pgvector
For the system to understand that “REST APIs” in a CV and “REST API design” in a posting mean the same thing, text is turned into vector representations. Instead of a separate vector database (Pinecone, Weaviate) I keep the vectors next to the main data: one database, one backup, less infrastructure to run.
Redis
Generating a tailored CV takes 10–15 seconds and costs tokens. Results are cached under a hash of their inputs, so an identical repeat request returns instantly — and the moment the CV or the posting changes, the hash changes too, so the cache never needs clearing by hand.
Spring AI
Claude handles the heavy text work, OpenAI produces the embeddings. A single interface means a model can be swapped for a cheaper or better one without rewriting the application logic.
Apache Tika
CVs arrive as PDF and Word. Tika extracts text from both through one tool instead of two separate libraries.
Testcontainers
372 automated tests. The integration tests run against a real PostgreSQL in Docker rather than an in-memory stand-in — slower, but it verifies how the system actually behaves.

Frontend

React 19 + TypeScript + Vite. Card dragging uses dnd-kit (react-beautiful-dnd is no longer maintained). TanStack Query caches server data and gives optimistic updates: a card moves to another column immediately, and if the server returns an error the change rolls back. The interface is built with shadcn/ui and Tailwind CSS.

Infrastructure

Backend on Railway, frontend on Vercel, PostgreSQL on Neon, Redis on Upstash. New versions deploy automatically on push.

Chrome extension · Backend

Subtitle Translator

Java 21Spring BootPostgreSQLTypeScriptChrome MV3Docker

About the project

Built entirely on my own: the browser extension, the backend, the database and the cloud deployment. The idea came from a problem anyone learning a language through video runs into — you either keep pausing to look words up, or you watch with a translation and remember almost nothing.

What it does

Translation inside subtitles
One click on a word shows the translation over the video, using the context of the line, so it is more accurate than a plain dictionary lookup.
Personal dictionary
Saved words, learning history, and sync across devices via Google sign-in. The extension also works without an account.
Spaced repetition
Words come back exactly when you start forgetting them: the interval grows after each successful recall, and difficult words show up more often.
Progress and video analysis
Statistics on what you have learned, highlighting of vocabulary you already know, and an estimate of how well a video fits your current level.

Technical decisions

TypeScript, no UI framework
The extension lives inside the YouTube page, so the main requirement is to stay out of the way. Plain TypeScript instead of React meant a smaller bundle, a faster start and no extra load on the page.
Chrome Manifest V3
Written against the current Chrome security standard from the start, so there is no forced rewrite when older extensions get migrated.
Java 21 + Spring Boot
The backend handles accounts, authorization, learning history and sync — here the priority is stability and an architecture that stays readable.
PostgreSQL
The data is tightly related: users, dictionaries, review history. A relational database guarantees integrity and cascades deletion of everything belonging to a user — a privacy requirement, not a convenience.
Automatic migrations
The schema changed several times during development; migrations run on every deployment, so the code and the data structure never drift apart.
Tests against a real database
Integration tests run on an actual PostgreSQL rather than an imitation — slower, but bugs surface before production.

Security

Encrypted keys
Users can plug in their own AI services through personal API keys, so those are stored encrypted: access to the database alone is not enough to use them.
Authorization
Separate access and refresh tokens, plus protection against password guessing and excessive requests to the auth endpoints.

Multilingual interface

The interface is available in Ukrainian, English and Polish, switching instantly without a reload. Plural and inflection rules were handled deliberately so the text reads naturally in each language.

Infrastructure

Managed cloud services for the database and the backend, with backups and security updates handled for me; new versions deploy automatically on a code change.