AI systems & implementation

I build AI systems you can inspect, trust, and put to work.

From serverless agents to grounded analytics, I connect real-shaped records, constrain model behavior in code, and design review paths that keep people in control.

  • Agents
  • Tool use
  • Integrations
  • Review systems

Selected work

Built like it has to hold up.

Three anonymized portfolio rebuilds based on real customer-success and support workflows. The company details and data are fictional; the architecture, implementation choices, and operating constraints are not. See every project →

Profile

Judgment at the boundaries.

Philadelphia, PA / building on the web

I'm Greg Finin. I like building useful software at the boundary between messy real-world records, operating decisions, and systems that need to be trusted before they are automated.

The portfolio projects here are anonymized rebuilds of real workflow and product patterns, using fictional companies and fixture data. They are built to show engineering judgment: clear data contracts, persistence, tests, deployable service shapes, grounded tool use, and review states.

The through-line is simple: connect the records, make the system behavior inspectable, validate structured outputs in code, and keep a human review path wherever AI is involved.

I'm currently talking with researchers and small organizations about the constraints they face in their data-intensive work.

GroundAnswers point back to inspectable source records.
ValidateModel output is untrusted until code checks it.
ReviewPeople keep the final say where judgment matters.