Service
I help companies find the task where AI saves the most time, choose it as the first project and put it into production. We define the project on a free scoping call, and a working first version is live 10 days later. The method comes from enterprise work: at a major Italian bank, an LLM agent routes more than 1 million help desk tickets a year with 81% accuracy against a 75% human baseline.
Most companies have more AI ideas than time to try them. The ones that pay off are internal, high volume and easy to measure: sorting requests, reading paperwork, checking data, answering the same questions every day. I look at how your team works today and find the task where AI gives back the most hours.
We settle it on the free scoping call: what to build, for whom and what done looks like. From there, a working first version is live in 10 days.
I learned this work in banking, where every system has to pass a security review. You get the same habits: models running in an EU cloud (Azure OpenAI in your own tenant when needed), no sensitive data in logs, an audit trail for every decision and a person in the loop. That makes GDPR and the AI Act much easier to deal with than adding controls at the end.
Azure OpenAI, Azure Document Intelligence, Azure Functions, LangChain, LangGraph, FastAPI, Kafka, OpenShift, Kubernetes, PostgreSQL, MongoDB.
I spent three and a half years at Sprint Reply on projects like these, for Italian banks and financial services clients. Client names are confidential.
In internal, high-volume work you can already measure: routing requests, checking paperwork, extracting data. At a major Italian bank that meant help desk routing (1M+ tickets a year, 81% accuracy), inheritance document checks (−80% handling time) and financial statement extraction (2,500+ documents a day). All three keep a person in the loop for exceptions.
Pick the task with the most volume, a number you already track and a low cost of error. If there is no baseline yet, measuring one is the first step. We decide it together on the free scoping call.
Yes, with the right setup. In my banking projects models ran on Azure OpenAI inside the client's EU cloud, sensitive data never went into logs and every production document had an audit trail. Some pipelines ran entirely on the client's own OpenShift.
A system in production, with results measured against the process it replaced. In each project I coordinated the development team, was the client's technical contact and built the architecture. I work in Italian and English.
Against your current process, with metrics you already track. In banking that meant accuracy against human routing (81% vs 75%), resolution time (5–6 days to 2–3), handling time (−80%) and manual work (−84%).
Related
Next step
Not sure where AI would pay off in your company? We can find the first project together.
Tell me the workflow. We scope it on a free call, then a working version is live within 10 days.
Book a scoping call