MSc Information Systems @ UZH · Software Engineering · Agentic AI · Data Science & ML · Knowledge Graphs
Patrik
Valach
I build agentic AI, data pipelines, and knowledge-graph systems. Software engineer at askEarth in Zurich, where I lead client projects end to end, turning data into AI assistants for government and industry clients. Previously co-founded an investor-backed fintech startup and co-authored published research on knowledge graphs.
Patrik Valach3+ yrsbuilding production software, from enterprise data teams to geospatial AI
~10full-stack & AI projects shipped end-to-end
1co-authored knowledge-graph research paper
4languages spoken fluently across the EU
Experience
Professional experience
07/2026 – PresentaskEarth AGZurich, CH
Software Engineer
- Lead client projects end-to-end, integrating and processing geospatial datasets and satellite imagery across on-premise and cloud deployments.
- Own the data pipeline design, turning raw geo data into AI-ready formats (PostGIS, GeoPandas, OpenSearch vector indices) for agentic pipelines.
- Build and equip AI agents with the right tool integrations using LangGraph, MCP and Azure OpenAI, deployed multi-tenant to Azure Container Apps via GitHub Actions.
- Develop and refine the UI/UX for both the general public and expert users across government, utilities, insurance and compliance.
- Technical and business point of contact for clients: reports, presentations, and supporting further projects across the team.
07/2024 – 05/2026Bora IT SolutionsFrankfurt am Main, DE
Software Engineer · Co-founder
- Co-founder of a B2B fintech startup venture.
- Leader of software development.
- Secured pre-seed investment.
- Development of custom fintech systems, vision was to use our investor-entrusted projects as base running revenue and demonstration track record to then attract new customers from the Frankfurt financial hub, focusing on certifications, reliability and long life-cycle that would foster client retention and satisfaction.
- Investor-supported project included architecting and development of a digital wallet and prepaid card payment system with physical POS vendor support and cryptocurrency integration.
- Wound down following co-founder's pivot.
07/2023 – 07/2024Lidl Digital InternationalData Enablement · Neckarsulm, DE
Werkstudent · Data Scientist
- Team-designed and ran an internal AI hackathon centred on knowledge graph exploration, generating 9 project proposals across 3 participating departments with 40 participants.
- Engineered end-to-end AI prompt pipelines and complex regex transformations for a media data architecture processing 50k+ product records, reducing manual transformation effort by ~97%.
- Benchmarked 21 transformer-based text and image embedding models (Sentence-BERT, CLIP variants, …) for cross-article mapping, improving match accuracy by ~56% vs. rule-based system.
- Partnered with 2 business departments to define data KPIs and produce technical documentation, enabling self-serve analytics for non-technical stakeholders.
Education
08/2025 – PresentUniversity of ZurichZurich, CH
MSc Informatics
Major in Information Systems, minor in Informatics. Expected 08/2027. Current grade 5.3iSwiss scale · 6.0 best · 4.0 to pass.
10/2022 – 09/2025TU MunichHeilbronn, DE
BSc Information Engineering
Computer Science and Management, focus on Data Engineering, LLMs, and ML. Grade 2.2iGerman scale · 1.0 best · 4.0 to pass, ranked 7th of 51 in the cohort.
Erasmus+ semester at the Arctic University of Tromsø (UiT), Norway.
2017 – 2022Evangelical LyceumBratislava, SK
English Bilingual Gymnasium
Exchange year at Bundesrealgymnasium Neusiedl am See, Austria.
Publication
Research
Q-NL Verifier: Leveraging Synthetic Data for Robust Knowledge Graph Question Answering
2025 · TU Munich, Professorship of Data Engineering · Co-authored, volunteer research
- Built a verifier for natural-language-to-SPARQL translations, trained on synthetic LLM-generated data, for robust question answering over knowledge graphs.
- Designed a reverse-translation LLM pipeline that generates NL↔SQL training pairs from human examples.
Selected projects
MSc project · full-stack digital therapeutics
- Integrating wearable sensor data with phase-aligned, interactive health dashboards.
- Built an MCP-based AI assistant so patients can query their own health data through an agentic interface.
- Running end-to-end user studies on usability and clinical utility.
- Python
- FastAPI
- React
- MCP
- Wearable APIs
- Local AI
- Tailwind CSS
- Pydantic
- MongoDB
- pytest
- Jest
- Claude Code
swisstopo · conversational access to federal geodata
- Open-source prototype under the Swiss Geoinformation Strategy, hosted by swisstopo: discovering and querying official geodata in natural language, without GIS expertise.
- Built askEarth’s work package — the agent backend, the MCP client for the geodata tools, and a versioned WebSocket protocol whose emitted frames are validated against the published JSON schemas in the tests.
- Wrote the evaluation harness, which doubles as a side-by-side benchmark of the two pilot models.
- Python
- MCP
- LLMs
- Amazon Bedrock
- Pydantic
- pytest
- NumPy
Event Crowdfunding DApp
2026UZH · Blockchains & Overlay Networks
- Crowdfunding platform on Ethereum Sepolia with ERC20 mechanics and a multi-contract architecture.
- Solidity contracts plus an off-chain oracle for deadline enforcement.
- Live with 35+ CI/CD releases on Vercel, IPFS for storage.
- Solidity
- ethers.js
- IPFS
- Ethereum
- React
- CI/CD
- Node.js
UZH · Network Science
- Scraped 16,400+ articles from NZZ.ch and ZEIT.de with full author and relational metadata.
- Built multi-layer co-authorship and citation graphs in Supabase.
- Used Louvain clustering, centrality, and assortativity to surface community structure and editorial organisation.
- Python
- NetworkX
- Selenium
- Supabase
- Plotly
- SQLAlchemy
- PostgreSQL
- Scikit-learn
- NumPy
SEET Match
2025Crisis Apps Hackathon · mentor matching
- Designed the architecture and a 5-dimension scoring pipeline (academic, language, geography, age, gender).
- Matches refugee students with Swiss mentors for the SEET NGO.
- TypeScript
- Python
- React
- REST API
- Optimisation
- Node.js
Legal Vantage
2024Graph-RAG legal Q&A system
- Graph-based RAG over legal documents, combining a Neo4j graph with a Nano vector store.
- Python API backend with agentic generation and semantic indexing.
- LightRAG
- Neo4j
- FastAPI
- Graph-RAG
- Docker
- Pydantic
- PyTorch
- Ollama
TUM · Data Engineering practical
- Generates natural-language training data for query models by running the translation backwards over the 24k-example LC-QuAD dataset — structured queries are plentiful, their natural-language counterparts are not.
- Llama 3 through Ollama does the verbalization, with Wikidata entity and property ids remapped to their natural-language forms first.
- Compared four evaluation metrics against manual judgements to fix the threshold that filters incorrect translations automatically, and fine-tuned a MiniLM embedding model for the distance measure.
- Python
- LLMs
- SPARQL
- Ollama
- SentenceTransformers
- PyTorch
- Scikit-learn
- SQL
Technical skills & tools
Programming
- Python
- TypeScript / JavaScript
- SQL
- SPARQL
- Solidity
Web & Backend
- React
- Next.js
- Tailwind CSS
- TanStack Query
- Zod
- FastAPI
- Pydantic
- SQLAlchemy
- Node.js
AI / Agent Stack
- LLMs
- LangGraph
- LangChain
- MCP
- RAG
- Azure OpenAI
- LangSmith
- Ollama
- Cursor
- Claude Code
Databases & Search
- PostgreSQL
- PostGIS
- OpenSearch
- pgvector
- Neo4j
- MongoDB
- Snowflake
- Supabase
ML & Data
- PyTorch
- TensorFlow
- PySpark
- Scikit-learn
- Pandas
- GeoPandas
- NumPy
- Arrow / Parquet
- SentenceTransformers
Cloud & Infrastructure
- Azure (Container Apps, ACR, Key Vault)
- Docker
- GitHub Actions
- Git
- pytest
- Jest
- Playwright
- Databricks
- Kafka
Management
- Scrum
- Jira
- Project Management
- Technical Documentation
- Stakeholder Management
Languages
- EnglishFluent · C1
- GermanFluent · C1
- SlovakNative
- CzechFluent · C1
Contact