Building with language models — orchestration, retrieval, protocol tooling, and spec-driven agent workflows.

• Agentic knowledge-base design (AGENTS.md)

  Structured context systems that let agents navigate, not rediscover.

  advanced(3 of 4)

  • → Senior Software Developer, RDI Software (Capgemini)

• GitHub Copilot (agentic use)

  Driving Copilot agents against a curated knowledge base.

  advanced(3 of 4)

• LLM orchestration & multi-agent workflows

  Planning agent plus implementation/test/review subagents.

  advanced(3 of 4)

• Prompt & spec engineering for agents

  Detailed spec writing that drives reliable agent execution.

  advanced(3 of 4)

• Embeddings & vector databases

  Embedding models and vector stores for retrieval.

  intermediate(2 of 4)

  • → MemoryHub

• LLM API integration

  Structured LLM calls in application pipelines.

  intermediate(2 of 4)

  • → VideoLearningTool - video knowledge triage

• MCP server development

  Building Model Context Protocol servers with guarded tools.

  intermediate(2 of 4)

  • → MemoryHub

• RAG & vector search

  Retrieval-augmented generation with embeddings and vector stores.

  intermediate(2 of 4)

  • → MemoryHub

• LLM fine-tuning (LoRA/QLoRA)

  Parameter-efficient fine-tuning of open models.

  beginner(1 of 4)

• LLM inference & serving

  vLLM, TensorRT-LLM, Triton, NIM - the target edge zone.

  beginner(1 of 4)

Models from the ground up — training loops, transformer internals, and the GPU compute beneath them.

• Experiment tracking (W&B)

  Tracking training runs and metrics.

  intermediate(2 of 4)

  • → WeightLifting - transformer from scratch

• Model training engineering

  Training loops, AMP, checkpointing, resume.

  intermediate(2 of 4)

  • → WeightLifting - transformer from scratch

• PyTorch

  Framework fluency from building and training models from scratch.

  intermediate(2 of 4)

  • → WeightLifting - transformer from scratch

• Transformer internals

  Attention, positional encoding, decoder-only architectures.

  intermediate(2 of 4)

  • → WeightLifting - transformer from scratch

• GPU & CUDA literacy

  GPU execution/memory model; entry-level CUDA.

  beginner(1 of 4)

Production services — distributed, event-driven backends and the cloud they run on.

• Python

  10+ years across scientific computing and commercial backend.

  expert(4 of 4)

• Async Python

  asyncio, tasks, and structured concurrency in production services.

  advanced(3 of 4)

  • → MemoryHub

• AWS (Lambda, DynamoDB, SQS, EKS, ElastiCache)

  End-to-end production AWS backend stack.

  advanced(3 of 4)

• CQRS & event-driven architecture

  Command/query separation and messaging.

  advanced(3 of 4)

• C# and .NET

  Primary commercial backend language; 5 years production.

  advanced(3 of 4)

  • → Senior Software Developer, RDI Software (Capgemini)

• Microservices architecture

  Event-driven, distributed backend systems.

  advanced(3 of 4)

• REST API design

  Designing REST APIs and gateways for microservices.

  advanced(3 of 4)

• SOLID, TDD & clean architecture

  Disciplined OO design and test-first development.

  advanced(3 of 4)

  • → Test-Driven Development (LearnQuest)

• Apache Kafka

  Event streaming in distributed backends.

  intermediate(2 of 4)

Shipping and operating software — containers, orchestration, pipelines, and infrastructure as code.

• CI/CD pipelines (Jenkins)

  Build and release automation.

  advanced(3 of 4)

• Code quality & security tooling (SonarQube, Snyk)

  Static analysis and dependency security in CI.

  advanced(3 of 4)

• Docker & containerization

  Containerizing services for orchestrated deployment.

  advanced(3 of 4)

• Kubernetes (EKS, Helm)

  Container orchestration and packaging in production.

  advanced(3 of 4)

• Shell scripting (bash, tcsh, PowerShell)

  Extensive CERN-era shell automation.

  advanced(3 of 4)

• GitOps with ArgoCD

  Declarative continuous delivery to Kubernetes.

  intermediate(2 of 4)

• Terraform (IaC)

  Infrastructure as code for cloud environments.

  intermediate(2 of 4)

Turning large datasets into answers — storage, querying, and statistical inference.

• Statistical analysis

  Frequentist methods, CLs, exclusion limits.

  expert(4 of 4)

  • → Single-lepton SUSY search (world-best limits in LHC Run 1)

• Large-scale data analysis

  Petabyte-scale experimental data pipelines.

  advanced(3 of 4)

• SQL & databases

  Relational and NoSQL data stores in production.

  advanced(3 of 4)

Scientific computing built for scale — analysis frameworks, simulation, and numerical precision.

• Distributed computing at scale

  WLCG: tens of thousands of cores, petabyte datasets.

  expert(4 of 4)

  • → Research Assistant, MTA Atomki

• C++

  10 years of scientific C++; wrote full analysis frameworks solo.

  advanced(3 of 4)

  • → Single-lepton SUSY search (world-best limits in LHC Run 1)

• Numerical methods & precision

  Physics numerical background; precision formats.

  advanced(3 of 4)

• Scientific computing (ROOT, GEANT4)

  CERN analysis frameworks and simulation.

  advanced(3 of 4)

Coordinating teams and explaining hard things clearly — across research and industry.

• Self-directed learning

  Career pivots and new-tech adoption without formal training.

  expert(4 of 4)

• Technical communication & presentation

  100+ scientific talks; clear complex-topic delivery.

  expert(4 of 4)

  • → Research Assistant, MTA Atomki

• Agile & Scrum delivery

  Working in Scrum teams with architecture influence.

  advanced(3 of 4)

• Mentoring

  Bringing colleagues up to speed, especially on AI tooling.

  advanced(3 of 4)

• Languages: Hungarian, English, German

  Hungarian native; English fluent (Advanced certs, full degree and career in English); basic German.

  advanced(3 of 4)

• Technical leadership & coordination

  L3 coordinator, group leader, team-lead track.

  advanced(3 of 4)

  • → PhD Research Associate & Group Leader, MTA Atomki
  • → Research Assistant, MTA Atomki