Implemented the guided diagnosis and briefing flow as a web MVP
← All work
AI assisted trades workflow
kleinagent.de
A web MVP that turns unstructured problem descriptions into a traceable brief and routes it to suitable trades businesses. The focus is controlled AI use and testable business logic.
- Role
- Product, business logic and full stack implementation
- Context
- Independent product
- Ownership
- Sole owner, with implementation and reviews orchestrated across specialised agents
- Delivery state
- Web MVP in development
- Evidenced
- Web MVP and documented test run evidenced
- Next proof
- Documented test state, market validation pending
- Period
- Since June 2026
Structured diagnosis in the implemented demo
01
Context
Customers often do not know which trade they need or what information is missing for a quote. Trades businesses simultaneously receive many incomplete or unsuitable enquiries.
02
My contribution
Modelled geo matching and lead acceptance as testable business logic
Modelled and tested payment, refund, webhook and idempotency logic against mocks
Added fraud and failure paths with fail closed behaviour
Implemented preview with example data and disabled actions
03
Decisions that shaped the product
AI structures, rules decide
The model assists diagnosis and structuring. Matching, payment states and permissions follow deterministic logic and remain testable.
Testable without credentials
External clients are loaded lazily and fail safely without keys. Core logic runs against PGlite and mocks so builds and tests do not depend on production access.
Review hardens payment logic
A holistic review found two race windows for duplicate charges or refunds in the prepared orchestration. I reserved state before the planned external call and made completion idempotent. The live Stripe integration remains a launch gate.
04
Technical evidence
Documented test run
302 tests passed- Source
- docs/handoff-2026-06-28-fast-life-8-13.md
- Verified
- 2026-07-16
Delivery process
6 delivery runs + review- Source
- docs/handoff-2026-06-28-fast-life-8-13.md
- Verified
- 2026-06-28
Verified revision
fbd81e4- Source
- local Git HEAD
- Verified
- 2026-07-16
05
Honest boundary
The screenshots show implemented demo components with synthetic data. They do not evidence a credential free interactive AI run or a database backed lead acceptance. The live Stripe LeadCharger and Setup Intent onboarding are missing, so real charges are not possible. A native Flutter app described in older documents does not exist yet either.