Jakub Kierznowski

• Java and Spring on Data Lineage: optimise performance over large enterprise metadata graphs by implementing graph and tree traversal algorithms. (Client data and exact figures are under NDA.)
• Cut the pipeline test run from 20 minutes to 4 by making the automated pytest suite asynchronous — parametrised cases, session-scoped fixtures.
• Own the Jenkins job scheduling for the pipeline.
• Daily Scrum in an English-speaking cross-functional team.

Council member since Mar 2024, leading its IT team since Mar 2025. I maintain the council website (Vue, TailwindCSS) and the VPS it runs on, and built internal automations on self-hosted n8n against Slack and Google APIs over OAuth 2.0. Event coordinator for the VOLT 2025 tech conference, eight partner companies including Aptiv, PwC Polska and Cytiva, and part of the organising team for VOLT 2026.

Mentored 35+ student teams per edition, from planning to the final presentation of their social-impact projects. Later ran a participant community (ZWzT × PwC certificate) and supported olympiad entrants.

Gameplay mechanics in Unity and C# for “Student Trainer”, a game built by a student team under a mentor.

• Native KSeF integration: API v2 against the FA(3) schema, KSeF-token auth on web, XAdES-BES certificate auth on desktop — filed straight to the Ministry of Finance, without a third-party integrator.
• A web PWA and a Windows desktop client share one account. PostgreSQL in the cloud, SQLite locally so the desktop client keeps working offline.
• Per-tenant isolation, KSeF tokens encrypted at rest. Stripe billing over card, BLIK and Przelewy24, a 14-day trial without a card, renewal flows, transactional email, PDF generation, XLSX and CSV import and export.
• Built and supported solo, including user contact.

An LLM agent with four retrieval tools over the Hetionet biomedical graph (47,031 nodes, 293k relationships): three Cypher queries and a similarity search, driven by a system prompt, walking drug → gene → pathway → disease paths to explain how a drug works and where it might be repurposed. Educational project on public research data. Demo at qualv13.github.io/neo4j-agent.

A simulation environment where agents negotiate power distribution across a grid under load. Agents trained with Proximal Policy Optimization in PyTorch and served back into the Unity simulation through ONNX. Two-person project: I gathered the data, built the environment and ran the training.

Compatibility is derived by traversal (device → shared port → machine → OS) rather than stored, so a search returns only hardware that actually connects, each hit with the path behind it. Gemini tool calling for semantic search and an inventory auditor, with deterministic fallbacks. A golden query set scores the graph engine against the flat one; 72 pytest tests on CI.

Builds playlists from a target BPM rather than from genre. Nobody else in the biohacking track approached it that way. Team of four, 24 hours, 3.9/5 from the jury.

Graph neural networks (GCN, GAT, GraphSAGE, GIN) over the Yelp dataset, benchmarked against one another. Transfer learning with Grad-CAM, DistilBERT sentiment classification, and network compression across seven architectures.

GPA 4.74/5.00. Year Representative since the 1st semester.

C++, competitive-programming problem solving.