AI solutions & automation
Applying large language models and agent-based systems to work that currently consumes people's days. The interesting problems are rarely the model itself — they are the plumbing around it: getting reliable input, constraining output, handling the cases where the model is wrong, and fitting the whole thing into an existing process without breaking it.
I build against your own data and systems, with attention to what happens when a result is uncertain. An assistant that is confidently wrong is worse than no assistant.
Typical deliverables
- Document, e-mail and order processing — extraction, classification and routing
- Question-answering over internal documentation and data (RAG)
- Agent-based tooling that operates against internal systems and APIs
- Decision support and anomaly flagging in operational data
- Proof of concept with an honest assessment of whether it should go further
