Berlin, Germany
Software Engineer with 7 years of experience. Now focused on AI solutions in Python and Golang. Passionate about building effective AI tools and platforms, both internal and public-facing. Background in backend engineering, DevOps, and DevRel across different domains. Open to roles where I can work on meaningful systems at scale.
- Delivered an end-to-end multi-agent RAG workflow for game content & internal knowledge automation, boosting query auto-resolution to 62% and cutting content generation cycle time by 45%. - Engineered a scalable multimodal ingestion pipeline processing 25K+ design docs, game bibles, and localized assets using Docling/PyPDF with hierarchical chunking, OCR, and rich metadata extraction. - Built a hybrid ReAct agent architecture utilizing Claude API & Model Context Protocol (MCP) with PostgreSQL/Go backend tooling; implemented strict slot-filling, deterministic tool calling, and HITL (human-in-the-loop) escalation paths. - Implemented full LLM observability and automated Evals suite via Langfuse, tracking latency bottlenecks, hallucination benchmarks (<1.5%), and token costs - reducing monthly inference spend by 28%. - Led canary rollout across internal studios (from 10% to 100% traffic) with Prometheus/Grafana SLO tracking, maintaining <800ms p95 retrieval latency and zero downtime.
Real-time highload app: latency-critical hot path, AI services over an internal knowledge base. AWS & GCP infrastructure. - Led migration of a legacy production system from local file storage to GCP, ensuring zero downtime and complete data integrity. - Built a RAG service grounding LLM answers in internal docs and a metadata catalog: structure-aware chunking, hybrid retrieval (pgvector + full-text, reranking), mandatory source attribution; plus a self-hosted inference layer over an OpenAI-compatible API - one codebase for local runners and production vLLM - Maintained IaC and GitOps pipelines across AWS (EKS, RDS, S3, SQS) and GCP using Terraform, Helm, and ArgoCD.
Architected and launched a next-gen microservices backend platform in Go and TypeScript (NestJS), replacing the legacy monolith and enabling rapid feature development. - Authored the initial draft, then led a team of 5 engineers in the process of adoption and migrating existing features. - Built an internal SDK that let frontend engineers self-serve backend logic, cutting coordination overhead and doubling delivery speed (~2x faster TTM). - Engineered a robust release pipeline powering the product’s first public launch (branching, e2e tests, AWS, monitoring).
Took over ownership of internal tooling, developer support, and cross-company DevOps coordination, freeing the CTO to focus on strategy. - Negotiated with designers and engineers to standardize asset validation checks, eliminating recurring issues and saving ~5 engineer-hours per week. - Extended internal static-checks framework, enabling workflows previously deemed impossible, such as running CI jobs locally after artifact download. - Delivered Python-based automation improvements, including localization workflows, spreadsheet-to-code pipelines, crash reporting, and build/config validation hooks.
As a freelance contractor, being an early WebAssembly adopter and enthusiast, I played a crucial role in shaping the DataDocs web spreadsheet software (Google Spreadsheets competitor): - Implemented a homegrown query engine, capable of Unicode characters matching, partial searches, and type coercion. - Boosted query performance in the app by x15, enabling full-featured 1 million rows search in less than 1 second. - Built an internal Python SDK for our product analysts and managers, enabling smooth experimentation with the query engine and algorithms, reducing hypothesis testing time from days to minutes.