Senior Full-Stack Engineer · AI systems, evaluation & cloud infrastructure
Berlin, Germany
I build software across the interface, backend and infrastructure, with 6+ years of experience in web applications and machine learning. My recent work focuses on scientific AI: multi-agent workflows, retrieval, evaluation and model serving.
I take responsibility for architecture and delivery, including the parts that keep a system usable in production: persistent state, background jobs, recovery, observability and deployment.
| Project | Engineering focus |
|---|---|
| Jev RAG Benchmark | Reproducible Python comparisons of Jev 1.13, OpenJev and NVIDIA rerankers on XQuAD-EN and SciFact. Frozen candidate pools, nDCG, probability calibration and paired bootstrap confidence intervals. Leaderboard. |
| BioAgents | Multi-agent scientific research system I helped build at Bio Protocol, covering research planning, literature synthesis, data analysis and hypothesis generation. |
| InClave | Co-built a local Python/Ollama assistant with sandboxed code execution, network denial and resource limits. A shared engine serves a CLI and a Tauri app through JSON-RPC. |
| Meetwit | Meeting assistant with local Whisper transcription, BGE-M3 embeddings and SQLite vector search, packaged with a FastAPI sidecar and Tauri desktop app. |
| Grex | A Helmor-based desktop workspace I developed for coding agents, integrating provider CLIs, Git worktrees, diff review and terminals. |
Co-author of Rethinking the AI Scientist: Interactive Multi-Agent Workflows for Scientific Discovery, January 2026. The paper describes BioAgents and its BixBench evaluation.
| Area | Technologies |
|---|---|
| Languages | Python, TypeScript, JavaScript, SQL, Solidity, PHP |
| Frontend & desktop | React, Next.js, Tauri, Yjs |
| Backend & data | FastAPI, Node.js, NestJS, Bun, Laravel, PostgreSQL, Redis |
| Agents & retrieval | LangChain, MCP, tool calling, RAG, embeddings, reranking |
| Evaluation & ML | LangSmith, human evaluation, LLM-as-judge, TensorFlow, scikit-learn, NumPy, pandas |
| Model serving | vLLM, Ollama, Amazon Bedrock, Whisper |
| Distributed workflows | BullMQ, SQS, Redis Streams, WebSockets, SSE |
| Cloud & deployment | AWS, DigitalOcean, Docker Swarm, Kubernetes, Terraform, AWS CDK, CI/CD |
Site development: DEVELOPMENT.md