Pick the right project
Not every AI idea deserves a team. I test the value, the data, the risks, and the appetite for change before anyone writes code, so the effort goes where it pays off.
See how I chose this projectI turn AI ideas into working software people actually use.
I'm Roxana Saedi. To show how I work, I built one AI project end to end — an assistant that sorts insurance applications — and wrote down every step: the plan, the tests, the safety rules, and the rollout.
The AI is the easy part.
The hard part is picking the right project, making the result trustworthy, and helping people work differently. That is my job.
Build it so it can be trusted
I turn the idea into clear requirements, tests, and safety rules — then into working software, released in small, reversible steps.
See the requirements and testsGet it used
New software only pays off when people change how they work. Training, feedback, and support are planned with the product, not announced after it.
See the rollout planOne project, shown end to end.
A life insurer receives applications. Most are simple and could move fast; some need an expert underwriter. I built an assistant that reads each application and recommends a route — plus everything a real project needs around it: a plan, requirements, tests, safety rules, and a rollout plan.
The case study walks through it in nine short chapters: how I chose the use case, how I made the requirements testable, how risks are controlled, the build itself, and how it would reach real users.
Read the case study — 15 min- 12requirements, each with a pass-or-fail test
- 8fictional applications used as fixed test cases
- 24automated checks that run before any release
Try it yourself.
Pick one of eight fictional applicants and watch the assistant decide: move the application straight through, or refer it to a human underwriter. It explains its reasoning and shows its audit trail. Nothing you do here is collected or stored.
Try the demo — no signupAPPLICANTSofia BianchiTeacher · CHF 900'000
ROUTERefer to underwriterPending cardiac investigation. Human review required.
A recommendation, not a decision — a person stays in charge
Use the same tools.
The checklist tells you whether an AI idea is ready to build. The playbook is the full method with templates. The workshop applies it to your team's own use cases.
CONTACT
Working on an AI idea that needs to become real?
If you are working on AI delivery, transformation, or adoption — especially in a regulated setting — write to me.