Hybrid dense + lexical retrieval over 4M documents with sub-100ms p95 latency and a reranking stage that lifted answer accuracy by 23%.
- Python
- pgvector
- FastAPI
- Rerankers
View projectSource
Available for new work
Your NameAI Engineer
I build production systems that put large language models to work — retrieval pipelines, agents, and the evaluation harnesses that keep them honest.
- Years shipping ML
- 6+Years shipping ML
- Inferences served
- 20M+Inferences served
- Contributor
- OSSContributor
Remote · Europe
About
Research-adjacent, production-obsessed.
I work at the seam between research and production: taking models that behave well in a notebook and making them behave well at three in the morning under real traffic.
Most of my time goes to retrieval-augmented systems, agent tooling and evaluation — the unglamorous scaffolding that decides whether an AI product is trustworthy or merely impressive in a demo.
LLM systems
RAG pipelines, tool-calling agents, structured output and prompt evaluation at scale.
Applied ML
Fine-tuning, embeddings, ranking and recommendation models from dataset to deploy.
Infrastructure
Inference services, vector stores, observability and cost control in production.
Product sense
Shipping AI features people actually keep using after the novelty wears off.
Selected work
Things I have built and shipped.
A few projects that show how I think about models, data and the systems around them.
A deterministic replay framework for multi-step agents: golden traces, tool mocking and regression scoring wired into CI.
- TypeScript
- LLM-as-judge
- CI
- Observability
View projectSource
Streaming proxy in front of several model providers with request shaping, failover, caching and per-tenant budget enforcement.
- Go
- Redis
- Streaming
- Kubernetes
View project
Layout-aware extraction from scanned PDFs into typed records, with human-in-the-loop review for low-confidence fields.
- PyTorch
- OCR
- Structured output
View projectSource
Detects distribution shift in production embeddings and alerts before retrieval quality visibly degrades.
- Python
- Metrics
- Grafana
View project
Small library for versioning, testing and diffing prompts as code. Used by a handful of teams in production.
- TypeScript
- OSS
- DX
View projectSource
Stack
Tools I reach for.
The day-to-day kit, roughly in order of how often it is open on my screen.
Languages
- Python
- TypeScript
- Go
- SQL
- Rust
ML & AI
- PyTorch
- Transformers
- LangGraph
- RAG
- Fine-tuning
- Evals
Data
- Postgres
- pgvector
- DuckDB
- Kafka
- Airflow
Platform
- Docker
- Kubernetes
- AWS
- Terraform
- GitHub Actions
Contact
Let us build something.
Working on something ambitious with models? I read every message and reply to the ones that need a reply.
you@example.com