AI Engineer
CV
LLM Evaluation · Retrieval Engineering · RAG Systems · EU-Sovereign Architectures
Professional summary
AI Engineer with end-to-end ownership of a RAG platform in the public sector — from the ingestion pipeline through retrieval optimization to an evaluated AWS Bedrock deployment. My background as a data engineer (ETL, data warehousing, large-scale data) shapes a system-first, data-driven approach — optimized for reproducibility, measurability, and operations. I think in systems, data flows, and failure modes.
Experience
- Since 10/2025: designed and built the end-to-end RAG pipeline of an internal GenAI assistant — an AWS Bedrock Knowledge Base over ~800 documentation pages, developed in a pilot Scrum team in close alignment with product owners and architects. Deployed and in user-acceptance testing ahead of production rollout (Aug 2026).
- Hypothesized that Bedrock's default chunking would not hold for the content — confirmed at ~30% Recall@5 in early measurements — and replaced it with a content- and source-aware chunking strategy, reaching 78% Recall@5 (MRR 0.66) on the expert-validated gold set (n=56), each lever measured in isolation against a purpose-built evaluation harness: three test suites, a 5-point recall regression gate (CI-blocking), bootstrap confidence intervals, and automated stakeholder reporting.
- Built an EU-data-resident, agentic AI development environment in AWS SageMaker (opencode → AWS Bedrock via a LiteLLM proxy) — credential-free via IAM execution roles, automated single-file bootstrapping, self-healing lifecycle.
- Data-engineering foundation in the BI team (SAP HANA / data warehouse): end-to-end ownership of production batch ETL at the billion-record scale — monitoring, metadata, root-cause analysis, production deployments.
- Object-oriented backend development (Java, Python) in an international, product-oriented team; encryption-focused API applications (PostgreSQL, REST APIs, Vaultree Encryption API).
Earlier experience (2014–2021): project & event management (Vienna) and vehicle-inspection engineering (KÜS), prior to transitioning into tech.
Selected independent systems: a public EU-sovereign RAG service over the EU AI Act + GDPR, live at philip-vana.com (FastAPI, Qdrant, EU-only inference) · a spec-driven SAP HANA warehouse engine with lossless, byte-stable schema round-trips (453 tests) · an agentic coding environment with tiered model routing behind one guardrailed gateway.
Skills
Generative AI / RAG
- RAG
- AWS Bedrock (Knowledge Bases, Titan, Rerank)
- Vector search
- Embeddings
- Chunking
- Query expansion
- LLM evaluation (Recall@k, MRR, nDCG)
- LLMOps
Cloud & infrastructure
- AWS (Bedrock, SageMaker, S3, Lambda, DynamoDB)
- IAM
- AWS CDK
- CodeBuild
- LiteLLM
Data engineering
- ETL / ELT
- Data warehousing
- Data modeling
- Batch & chunk processing
- SAP HANA XSA
- Oracle DB
- PostgreSQL
Languages & tools
- Python
- SQL
- FastAPI
- Pytest
- Git
Education
Deutsche Rentenversicherung Bund, Berlin
2021–2024
Final grade: 1.7 (German scale, 1.0 = best)
Wilhelm-Busch-Realschule, Munich
2007–2012
Final grade: 2.0
Languages
- German (native)
- English (C1, fluent)