Project case study
BreakPoint
It is easy to get excited about an idea before testing what has to be true for it to work. BreakPoint gives that excitement some structure: define the venture, clarify the assumptions and look for the point where it breaks.
My role: Concept, system design and evaluation framework, with AI-assisted implementation.
Watch the interactive walkthroughHow it works
Describe the venture
A short idea becomes a structured request.
The API validates the idea, stage and supplied answers with Zod before making a model request and checks its configured daily quota.
Answer the hard questions
Clarifications make the assumptions explicit.
The evaluation combines the venture description and answers into the model input rather than treating each exchange as an unrelated prompt.
Examine the weakest assumption
The result organizes failure scenarios and missing evidence.
The server requests a schema-constrained response from OpenAI, normalizes the fields and checks narrative completeness.
Choose the next experiment
The output points to evidence the founder must gather.
The structured proof-required and strengthening sections turn the analysis into testable next steps. The model does not perform those experiments.
Find the assumption worth testing next.
I designed BreakPoint around a structured flow rather than an open-ended chat. The user describes an idea and supplies clarifying information. The result organizes the weak assumptions, potential failure scenarios and evidence needed before moving forward.
- Focus
- Next.js, TypeScript, Zod and OpenAI structured responses
- Status
- MVP; decision-support prototype
How it works
A Next.js API validates structured input with Zod, checks the configured quota and calls OpenAI for a schema-constrained analysis. The response is normalized and checked for complete narrative sections before the interface presents it.
A decision that mattered
The output has to be usable as a decision aid. A strict response structure makes missing sections detectable, while completeness checks reject an answer that looks formatted but does not contain the needed reasoning. The model’s verdict still needs human judgment.
Status
BreakPoint is an MVP for structuring the next experiment. Its output needs customer evidence and human judgment before it should change a real business decision.