I build systems around large language models.Ones that run in production.

Data extraction, classification and enrichment at scale.

Every result carries a confidence score, uncertain cases go to review, and nothing is written to the source system without confirmation. Most of the work is not in the prompt but in what happens before and after it: validation, duplicate detection, and a trail that shows where each record came from.

Assistants that answer questions from a company's own material and point to the source of every answer.

I work with OpenAI and Anthropic models, and where data cannot leave the company, with open models running on the company's own infrastructure. I spend more time on retrieval quality and on testing the answers than on the interface itself.

A weekly report, data reconciliation, moving information between systems.

An agent runs on a schedule, has a set budget, logs what it did, and lets you know when something fails. Any operation that changes something in an external system needs approval first.

A common starting point is a prototype that looked good in a demo but is not ready for daily work yet.

I keep what works, add tests, monitoring and error handling, and turn it into a service someone else can maintain.