01 // LIVE PRODUCT
todrawn.com
todrawn.com combines Scribe for hand-drawn video creation with collaborative boards; the board mode runs on a Java 21 / Spring Boot backend and a Next.js frontend.
SYSTEM STATUS: BACKEND + RAG
JAKUB KUŹNICKI
// brbn-jpg · Java Backend & RAG Developer · Warsaw, PL
I build backend systems and RAG products, shipping production work with Java and Spring Boot.
/ evidence
Selected proof
Three concrete signals from products and work already in the repo.
02 // RAG PIPELINES
Document ingestion
Built LangChain ingestion pipelines and RAG systems during a 2025 backend internship, including containerisation and deployment of LLM-driven processing.
03 // LOCAL LLM
LlamaTalks
A Spring Boot and LangChain4j chatbot using Ollama, document-based RAG, persistent chat history and SSE response streaming.
EXPERIENCE
Backend work across enterprise Java and LLM data pipelines.
Junior Java Developer
Netcompany
Working on enterprise-level Java software at Netcompany.
Backend Developer Intern
Paweł Weselak Pragmatyczne Programowanie
Built LangChain-based ingestion pipelines and implemented RAG systems using LangChain4j & Spring Boot. Managed containerization and deployed LLM-driven data processing solutions.
SELECTED PROJECTS
Read the problem, the engineering choices and what each project delivers.
Java, TypeScript
todrawn.com
- Problem
- For the board mode, keep drawing and board state in sync across connected clients while the wider product also supports Scribe, accounts and subscriptions.
- Role
- Built todrawn.com end to end; this case study focuses on the Spring Boot API, Next.js canvas experience, real-time layer, billing and deployment behind the board mode.
- Decisions
- WebSockets carry board updates; PostgreSQL stores durable data; Redis handles fast, short-lived state; Docker, GitHub Actions and Cloud Run handle delivery.
- Result
- A live todrawn.com product combining Scribe with collaborative boards, including a production board mode with Stripe subscriptions.
Java
LlamaTalks
- 01 Document Tika → chunks
- 02 Vector store embeddings + retrieval
- 03 Ollama context + history
- 04 SSE Reactor/Flux
LangChain4j orchestration · PostgreSQL history
- Problem
- Answer questions from a user's documents while keeping the language model local to the application.
- Role
- Built a Spring Boot chatbot around LangChain4j, Ollama, document ingestion and persistent conversations.
- Decisions
- The ingestion endpoint sends files through Apache Tika parsing and chunking, then embeddings and semantic retrieval provide context alongside conversation history. LangChain4j connects that context to Ollama, while Reactor/Flux and SSE stream the response; changing embedding dimensions requires re-ingestion.
- Result
- An open-source chatbot project with local-model conversations, document Q&A and streamed responses.
Java, React
Gamelog
- Problem
- Give players one place to organize a game library, track progress and keep social features beside the backlog.
- Role
- Built a full-stack game backlog manager with a Spring Boot API and React interface.
- Decisions
- A Kanban dashboard models progress; PostgreSQL persists accounts and library data; Redis provides caching; Docker packages the application.
- Result
- A working full-stack tracker with search, playtime, ratings, friend requests and a live demo.
Java, Vue, TypeScript
MapGenerator
- Problem
- Turn one seed and a map size into an explorable fantasy world while keeping large maps responsive in the browser.
- Role
- Built the Spring Boot generation backend and Vue/TypeScript Canvas client, including the streaming path, viewport and renderer.
- Decisions
- Multiple noise layers model height, moisture, temperature and the continent mask. The client uses a hybrid virtual/offscreen canvas: static land is pre-rendered and only waves are calculated per frame.
- Result
- A working procedural map generator with streamed terrain, pan/zoom navigation, topology shading, procedural textures and cities.
Java, Angular
Cibaria
- Problem
- Give recipe enthusiasts one place to create, store, search and share recipes with a bilingual interface.
- Role
- Built the full-stack application across its Angular frontend and Spring Boot REST API, connecting the product to PostgreSQL.
- Decisions
- Angular 18.2.7 provides the responsive client, Spring Boot 3.4.1 exposes the API, PostgreSQL persists recipe and user data, and the product keeps English and Polish content paths together.
- Result
- A live recipe manager with recipe creation, search, filtering, profiles, favorites, ratings and English/Polish support.
Astro, TypeScript
SosniakPortfolio
- Problem
- Present Sebastian Sośniak's projects and experience in a fast portfolio that stays easy to maintain as the content grows.
- Role
- Designed and built the portfolio with Astro, TypeScript and Tailwind CSS, then deployed it to the live metricmind.pl domain.
- Decisions
- Astro's mostly static output keeps the content path light; TypeScript supports type-safe components; Tailwind CSS keeps the visual system consistent; Vercel provides hosting and delivery.
- Result
- A live content-first portfolio for Sebastian Sośniak at metricmind.pl, backed by a public repository with 10 commits.
Astro, TypeScript
StrefaSA
- Problem
- Connect an MTA:SA community's game atmosphere with a credible web presence and a shop path that hands checkout to the separate backend.
- Role
- Worked as co-owner and co-creator of the server, designing and building the Astro frontend, including the marketing site and /sklep handoff.
- Decisions
- Astro keeps the public frontend lightweight; live server data builds trust; /sklep hands payment flow to a separate MTA backend and Stripe Checkout instead of pretending the frontend owns fulfillment.
- Result
- A live cinematic marketing site and store entry point at strefasa.pl, backed by a private repository with 26 commits.
/ open channel
CONTACT
Have a backend, AI or infrastructure problem worth solving? Send me a note.