Refinery · Agent Lab · Workshop

Services

Data pipelines

Raw data in, usable data out. I build ingestion and modelling pipelines that turn scattered APIs, spreadsheets and inboxes into data you can trust and act on, tested and version-controlled like any other software.

AI agents

Agentic workflows that do real work: research, enrichment, lead handling. They run on local inference where privacy matters, use retrieval over your own data, and come with evaluations that check what the model actually does.

Web apps

Practical tools for real business problems, built alongside the people who’ll use them, from internal dashboards to customer-profiling tools that pull several data sources together.

Security first

Every engagement starts from least privilege. Agents run in sandboxes, credentials stay with a broker rather than in prompts or code, and each integration can only reach the endpoints it needs.

Stack

• Languages: Python, SQL, JavaScript, Rust (→ WebAssembly)
• Data: dbt, Postgres, SQLite, Databricks, Weaviate
• AI: local LLM inference, RAG, MLflow-based evaluation
• Platform: Docker, Terraform, GitHub Actions, Azure, GCP