MemoryHub
Content-agnostic engine over a markdown memory store with schema-validated writes and an MCP server.
- Python
- Pydantic
- Typer
- FastMCP
- LanceDB
- pytest
Author (spec-driven, built with AI agents)GitHub
AI Engineering · Cloud Backend · Physics PhD
Dr. János Karancsi
AI engineer with a physics PhD and a decade at CERN — building agentic systems that let coding agents plan and execute senior-level work.
Currently — Senior Software Developer at RDI Software Hungary (Capgemini Group)
- 2010Physics · CERN
- 2021Cloud backend
- 2026AI engineering
Selected projects
Featured work
WeightLifting - transformer from scratch
Decoder-only char-level transformer (~10.8M params) built and trained from scratch in PyTorch.
- PyTorch
- CUDA
- Weights & Biases
Sole authorGitHub
Razor-boost SUSY search (PhD thesis, world's best limits in Run 2 for a period of time)
Boosted-object Supersymmetry search on 2016 LHC data using razor variables - the analysis behind my PhD thesis, coded, plotted, and statistically analysed in the BoostAnalyzer17 framework.
- C++
- Python
- ROOT
- WLCG
Main author of the analysis software for the Razor Boost analysisGitHub
Recent