Jan Bancerewicz / AI & software developerOpen to AI/ML / full-stack roles

I build software that
survives contact with

real data.

Nice '26Hover: run detection

I work on applied machine learning: LLM interpretability, computer vision, and model optimization – as well as the full-stack systems needed to carry them in production. Final-year CS student at Gdańsk Tech with over 1 year of commercial experience in IT, spending my free time competing at hackathons and conducting AI research.

See the work↓Get in touch↗

23+Projects shipped

6Hackathons entered

8+Certifications earned

1+Years of commercial xp

Gdańsk, Poland

PythonJavaSQLTypeScriptC++PyTorchHugging Facescikit-learnLoRA/PEFTMLflowPandasvLLMLlamaIndexRAG SystemsQdrantAnthropic/OpenAI APIGoogle ADKLLM InterpretabilityMultimodal / VLMsComputer VisionYOLOOpenCVSignal ProcessingModel OptimizationFastAPISpring BootPostgreSQLRedisMongoDBAndroidDockerKubernetesAzure DatabricksAWSCI/CD & GitHub ActionsReactNext.jsClaude CodeCursor IDE

01/Selected work

All 5 projects ↗

My flagship research and full-stack projects. Each breakdown highlights architectural decisions, real-world metrics, and key takeaways.

01 / LLM Engineering / 2026

LLM Internal States Monitoring↗

Full-stack project designed to turn black-box LLMs into fully observable systems. By running local checkpoints in-process and attaching forward hooks across the decoder stack, it computes logit lens projections, attention dynamics, and decision-depth metrics displayed in a React and FastAPI dashboard.

LLM Interpretability
PyTorch
FastAPI
React
Logit lens
RAG
Transformers

02 / Multimodal LLMs, Vision-Language / 2026

Qwen-VL for Angiography↗

Adapting Qwen3-VL for coronary vessel segmentation using DeepStack and LoRA, benchmarked against ResNet-34 U-Net on ARCADE SYNTAX. Extension of the Arterio labeling platform at Gdańsk University of Technology

Qwen3-VL
PyTorch
Hugging Face
LoRA/PEFT
DeepStack
Medical Imaging
Transformers
Segmentation
Arterio

03 / Research, Applied ML, Biosignals / 2025

Analysis of HRV Using Mobile Devices and Machine Learning↗

Research project for mobile ECG/PPG peak detection and HRV analysis. Compares 1D-CNN detectors against classical Pan–Tompkins/NeuroKit baselines using smartphone camera PPG and Polar H10 ECG. Peer-reviewed TASK Quarterly publication (March 2026).

PyTorch
Signal Processing
Android
Python
Research
HRV
PPG
ECG

02/Hackathons & competitions

6 programming challenges
2 podium finishes

How I enjoy spending my weekends:
a complex problem, something impactful to build
and zero time for overthinking.

Nov 2024 & 2025

AMPPZ – Polish Collegiate Programming Contest (2x)↗

Warsaw – University of Warsaw

Representing my university in the Polish collegiate programming contest (ICPC) since 2024, placing #34 nationally in 2024 and #53 in 2025 (out of ~80 teams). As President of the Sfera PG Algorithmic Club, I lead team training in advanced algorithms, complex data structures, and high-pressure problem-solving in C++.

Algorithms & Data Structures
ICPC
C++
Problem Solving

34th nationally in 2024

Representing Gdańsk University of Technology

Jan 2026

EU Critical Infrastructure Hackathon↗

Gdańsk – Gdańsk Science and Technology Park

Engineered a security layer for PV inverters, insulating critical grid infrastructure from remote cyberattacks through real-time anomaly detection for remote control. Evaluated and awarded 1st place by an international jury of industry experts.

Cybersecurity
Energy
Modbus32
Embedded
React Native

1st place

Energy & Offshore category

Nov 2025

Heroes of the Brain – Neurohackathon↗

Wrocław – Wrocław University of Science and Technology

Developed AuraCloud, a Brain-Computer Interface translating raw 8-channel EEG into a continuous three-dimensional emotional space (PAD model) and an automated stimulus path using OASIS dataset to actively guide user mood.

BCI
EEG
Signal Processing
Dash
Python

2nd place

Signal processing & modelling

May 2026

Hackology II – AI Hackathon↗

Lublin – Lublin University of Technology

Built a 369-class retail product detection pipeline evaluated on mAP@0.5. Combined YOLO and RF-DETR detection with embedding-based retrieval and multi-step reranking to identify products on dense store shelves.

Computer Vision
YOLO
RF-DETR
Machine Learning
Python

Shipped a full pipeline

Computer vision & Object Detection

Oct 2025

HackYeah↗

Kraków – Tauron Arena

My first hackathon! I built StressLess – a mobile biofeedback app measuring heart rate and HRV via smartphone camera PPG. Combined biometric signals with an LLM pipeline to deliver real-time stress analysis and biohacking insights.

Android
Signal Processing
LLM API
Java

Solo debut, high score

Biohacking Category

03/Internships & commercial experience

2 companies/15 months

Over a year of commercial experience in software engineering and high-throughput production systems, gained alongside studying Computer Science.

Jan 2026 — presentCurrent
9 months
Data & Software Engineering Intern
- Working in the Pricing department, I develop and maintain PricePoint, the insurance premium calculation system that takes over 200,000 requests a day.
- Designed and hosted cloud pipelines on Azure Databricks, developing a translator to convert Earnix risk formulas into scikit-learn prediction models.
- Delivered features through a CI/CD workflow, collaborating with the team on iterative deployments.
- Python
- SQL
- Azure
- Databricks
- Dynatrace
- Git
- Agile
Jul — Dec 2025
6 months
IT Department Intern — DevOps & systems administration
- Administered OT infrastructure across a fleet of 90 virtualized servers, managing VMs and system configurations on both Windows and Linux.
- Developed automation scripts in PowerShell and Bash to handle server monitoring, alerting, and backup processes.
- Gained hands-on experience with DevOps tasks: deployment, monitoring, and keeping systems up.
- Hyper-V
- PowerShell
- Bash
- Linux
- Windows
- Azure
- Vim

04/Certifications & courses

Academic foundation, published research, and specialized coursework targeting deep learning, LLMs and software engineering.

19 entries/7 sources

Gdańsk University of Technology

BSc in Computer Science (4.41 GPA) & leadership roles

Bachelor's thesis – defending Feb 2027
Gradient PG - AI & Machine Learning Science Club – Board Member
Sfera PG - Algorithmic & Competitive Programming Club – President

Research & programmes

Peer-reviewed publications & research initiatives

CORE: Comments as a Reasoning – Gradient PG Science Club (Paper in-progress)opens in a new tab
Analysis of HRV Using Mobile Devices and Machine Learning – TASK Quarterlyopens in a new tab
Europe AI Summer Research programme - participant (Edition in-progress)

Google

AI business applications & cloud

Umiejetnosci Jutra AI 3.0opens in a new tab

NVIDIA Deep Learning Institute

High-performance AI & GPU acceleration

Microsoft

Microsoft certifications

GitHub Foundationsopens in a new tab

DataCamp

Deep learning specialisations

Academic Partners

Modern AI Stack & Applied Software Architecture

05/Writing

All posts ↗

Notes on the technical topics that might interest you.

12 Aug 2026

How to cut your Claude Code token usage by 20x↗

Claude bills a cached token at 0.1x and a cache write at 2x – a 20x spread on identical content. How context and prompt caching work in Claude Code, the hygiene that cuts always-on context, and a one-plugin guard that blocks cold-cache turns before you pay.

LLM tooling/8 min

10 Aug 2026

How to debug Python like a pro↗

Debugging in PyCharm is not print-spam. Breakpoints freeze the process, Step Over/Into/Out move with intent, diffs show what changed, and one mute-all trick lets you jump without getting snagged.

Developer Tools/10 min

28 Jul 2026

What quantization really is↗

Quantization is not magic compression. It is a controlled change in numerical representation that trades precision for memory, bandwidth, and often speed – illustrated with Gemma 4 12B.

AI Engineering/8 min

06/Stack

39 tools

Technologies I reach for by default — tools I’ve trained models with, built pipelines around, and shipped to production.

Languages

Python
Java
SQL
TypeScript
C++

ML Frameworks & Training

PyTorch
Hugging Face
scikit-learn
LoRA/PEFT
MLflow
Pandas

LLM Serving & RAG

vLLM
LlamaIndex
RAG Systems
Qdrant
Anthropic/OpenAI API
Google ADK

AI Domains & Specializations

LLM Interpretability
Multimodal / VLMs
Computer Vision
YOLO
OpenCV
Signal Processing
Model Optimization

Backend & Databases

FastAPI
Spring Boot
PostgreSQL
Redis
MongoDB
Android

Cloud & DevOps

Docker
Kubernetes
Azure Databricks
AWS
CI/CD & GitHub Actions

Web & GenAI Tools

React
Next.js
Claude Code
Cursor IDE

07/Contact

If this matches what
you are looking for
– say hello.

LinkedIn or email, whichever you prefer – I read everything and reply within a day or two. Tell me what you are building and I will tell you honestly whether I am the right person for it.

Open to AI/ML / full-stack roles. Available for full-time roles. Based in Gdańsk, Poland, comfortable working remote across European time zones.

Message me on LinkedIn↗Email instead→

LinkedInjan-bancerewicz GitHubJanBancerewicz CVDownload PDF ORCID0009-0001-3898-5576

jan@portfolio: ~

Jan Bancerewicz — AI & software developer

An interactive link tree. Type `help` for the command list.

Try one

I build software thatsurvives contact with