[ 01 ]  Bratislava, Slovakia

JÁN

MICHÁLEK

AI Engineer  ·  ML Researcher  ·  Systems Thinker

Python Machine Learning FastAPI Qdrant Vector Search Arch Linux
View Projects Contact →
Scroll

About

Building intelligent
systems from
first principles.

2nd-year student at SPŠE Hálova, Bratislava — also attending university-level lectures at FEI STU via the Habsburg program. I build AI pipelines, research ML models, and ship real products. My work ranges from semantic search infrastructure to train-delay prediction at national scale.

3rd-place finish at Open Data Heroes Hackathon with an AI fact-checking assistant. Incoming intern at Asseco CEIT (May 2026), working on AGV systems and Industry 4.0 solutions.

Quick Facts

  • → Daily driver: Arch Linux + Hyprland
  • → CTF / Cybersecurity competitions
  • → Power metal listener
  • → Deep interest in math: complex numbers, linear algebra, calculus
  • → Attending CODECON 2026, Bratislava — May 14

Currently Working On

  • ZSR Train Delay Prediction — research paper target
  • Slovak Legal & Financial RAG Assistant
  • HuggingFace publishable ML projects

Projects

Selected Work

01 — Research

ŽSR Train Delay Prediction

In Progress · Paper Target

Two-stage LightGBM pipeline (binary classifier → regressor on delayed subset) over ~4.8M journey snapshots. Graph analysis of the Slovak rail network (385 nodes, 1018 edges), PyArrow batch processing, and Open-Meteo weather integration. Ground truth always derived from Cas − CasPlan.

LightGBM PyArrow Graph Analysis Open-Meteo MongoDB GTFS

02 — NLP / RAG

🥉 3rd Place

Demagog.sk Fact-Checker

AI fact-checking assistant for the Open Data Heroes Hackathon. BGE-M3 embeddings over 22k political statements, semantic search via Qdrant, RAG pipeline with tool-calling chatbot and Tavily web search integration.

BGE-M3 Qdrant FastAPI Vue.js Tavily

03 — RAG / Search

Slovak Legal & Financial Assistant

Hybrid search (BGE-M3 dense + sparse) with Qdrant RRF fusion, multilingual cross-encoder reranker, and qwen2.5:7b-instruct via Ollama. Trained on SK-QuAD (~22k deduplicated paragraphs).

BGE-M3 Qdrant Ollama Streamlit FastAPI

04 — Deep Learning

Neural Network from Scratch in C++

Fully hand-rolled neural network trained on MNIST — no ML frameworks. Implemented forward pass, backpropagation, cross-entropy loss, and softmax from first principles, building a rigorous foundational understanding of transformer architecture and attention mechanisms.

C++ Backpropagation MNIST Cross-Entropy Softmax

05 — Platform

SK Political Fact-Check Platform

Production evolution of the hackathon project. Modular monolith with verification_service, research_service, and qdrant_service. Adapter-pattern external DB integration.

FastAPI Qdrant Adapter Pattern Modular Arch

Stack

Technical Arsenal

Languages

Python Primary
C++ Proficient
Rust Learning
SQL Proficient

ML / AI

LightGBM / XGBoost
BGE-M3 Embeddings
Qdrant Vector DB
RAG Pipelines
Cross-encoder Reranking
Transformer Architecture
Neural Net from Scratch (C++)

Infra & Tools

FastAPI + SQLModel
Vue.js + TailwindCSS
Ollama (local LLMs)
Cloudflare Pages & Functions
uv (Python env)
MongoDB / PostgreSQL
Arch Linux + Hyprland

Mathematics

Complex Numbers Linear Algebra Calculus I–II Discrete Math Statistics Optimization Graph Theory Physics Simulations

Experience

Timeline

May 2026 — Upcoming

Intern — Asseco CEIT

AGV · Industry 4.0

Joining Asseco CEIT's team focused on AGV systems, industrial software, digital twins, and Industry 4.0 solutions. Connecting ML background to real-time industrial optimization.

2024 — Hackathon

Open Data Heroes — 3rd Place

🥉 Award

Built AI fact-checking assistant for Demagog.sk. Full semantic search and RAG pipeline over 22k political statements. Stack: Python, FastAPI, Qdrant, Vue.js.

Ongoing — University

FEI STU — Habsburg Program

Bratislava

Attending university-level lectures at the Faculty of Electrical Engineering and Information Technology, Slovak University of Technology.

Ongoing — Secondary School

SPŠE Hálova — Bratislava

2nd-year student at the Secondary Electrotechnical School. Focus on programming, electronics, and applied mathematics.