Stefan
Stradviso Decision Studiowith the real decision interface, human confirmation and Build Week presentation.

OpenAI Build Week 2026 / Work & Productivity

Stradviso Decision Studio

As a solo founder, I built a controlled AI decision system with Codex duringOpenAIBuild Week that converts customer questions into traceable project choices, without transferring human decision-making authority to the model.

Build Week entry Technical proof

The case in 30 seconds

Stradviso Decision Studio

Problem
As a solo founder, I built a controlled AI decision system with Codex duringOpenAIBuild Week that converts customer questions into traceable project choices, without transferring human decision-making authority to the model.
Solution
Growth platform with positioning, go-to-market, AI strategy and an interactive growth scan.
Role and focus
Strategy, processes
Status
Build Week Concept Lab
Result
Analysis, growth path

Live lab / try it yourself

Follow a supervised AI decision path

See how a customer request is defined, translated into traceable options and only allowed to continue in a controlled manner after human approval.

Open Decision Studio View the story
01 / InputCustomer demand, context and uncertaintyYou can change or choose this yourself
02 / SystemAI options, approval and allowlistOpenAIResponses API · Human approval · Audit trail
03 / EvidenceA traceable route with human directionThis makes the operation visible
PlatformBusiness OS + Decision StudioStrategy becomes a workable system
My roleProduct vision, AI governance and final directionFrom customer problem to verified release
AI laneOpenAIResponses API · GPT-5.6 SolPeople-driven and limited
StatusWorking R30 release submittedFurther product development continues

Product presentation

Decision Studio

This compact presentation lets the product itself speak: from a customer request and missing context to defined routes, human approval and controlled execution.

  • 01Question and context
  • 02Routes and human choice
  • 03Safe execution and control
View full Build Week entry

Build Week in numbers

Frozen release. Proven pace

Within 5 days and 15 hours of turnaround time, I built a complete Business OS and Decision Studio release together with Codex.

Lead time
5 days 15 hours
Repository
134 commits
Product scope
460 files
Test evidence
473 tests
Assertions
6.431
CI
SQLite + MySQL passed
Release
Immutable R30
Approval limit
3 records · 0 stale writes

The product question

A customer request is not yet a project decision

“Can we launch two weeks earlier without increasing the budget?”

The question sounds concrete, but capacity, delivery dates, ownership and consequences are not yet clear. Converting directly to a commitment would build uncertainty into the project.

  1. 01Context

    Only authorized project context is collected.

  2. 02Conflict

    Missing capacity and conflicting delivery dates become visible.

  3. 03Clarify

    The system first asks one targeted clarification question.

  4. 04Options

    GPT-5.6 then develops three defined routes.

  5. 05Authority

    A responsible person will choose the exact route.

  6. 06Apply

    Only Decision, Risk and Milestone change together.

  7. 07Recheck

    If the project has been changed, it will be rejected with zero writes.

01
AI proposesExplores restricted routes
02
A human decidesAssesses context and consequence
03
The system enforcesOnly performs what is permitted

Four steps. One monitored transition.

From analysis to growth path

The interface makes the decision status visible and keeps context, permission and execution separate.

QUESTION

Formulate what needs to be decided

Choice, desired timing and most important uncertainty are explicitly recorded.

OPTIONS

Compare restricted routes

AI provides three traceable options or stops when one missing piece of data needs to be answered first.

APPROVE

Tie approval to context

An authorized user reviews the chosen route and confirms the exact consequences.

RESULT

Write only what is allowed

Go-to-market and digitalization follow the same rule: the atomic mutation is limited to Decision, Risk and Milestone.

DECISION STUDIO / QUESTIONAUTHORIZED CONTEXTHUMAN CONTROL
Decision Studio home screen with a project question, timing, uncertainty and explicit confirmation for live AI.
Product proof / 01Live AI is only accessed after explicit confirmation of the demarcated fields.
Governed decision flowFrom question to traceable decision

Each step monitors context, human authority, and the exact records that are allowed to change.

  1. 01Customer questionTiming, budget or risk is made explicit.
  2. 02Authorized contextOnly the correct organizational and project data are included.
  3. 03Clarification or 3 routesMissing context first leads to one focused question.
  4. 04Human approvalAn authorized user selects and confirms the route.
  5. 05Revision checkThe approval is reassessed against the current context.
  6. 06Decision + Risk + MilestoneThe allowlist writes exactly three related records.
  7. 07Audit receiptA line of evidence makes the implementation traceable afterwards.

Governance as product behavior

Don't appear smarter, but work in a controllable manner

The most difficult design choice was not how the AI produces an answer, but what it should never do on its own.

01

Authorized context

The workflow assembles canonical project context within organizational and project permissions.

BOUND
02

Refusal before provider use

Stop sensitive or cross-organization requests before key lookup or provider access.

REFUSE
03

Human approval

An AI output cannot write directly; consent remains explicit and context-specific.

APPROVE
04

Atomic allowlist

Only Decision, Risk and Milestone are allowed to change. A legacy context writes zero records.

ENFORCE
05

Audit receipt

The implementation provides an append-only proof rule with which the permitted mutation can be verified.

TRACE

Claim limit: store:false is a request setting, not a universal zero-retention claim. Provider processing remains subject to the applicableOpenAIterms and account settings.

Examined jury evidence

Four scenarios make the borderline behavior visible

The jury route demonstrates four checkpoints with representative, synthetic test data: clarify, approve, block outdated context and safely reset. Changing management functions remain shielded in the public environment, so that visitors can reproducibly inspect behavior without changing the evidence status.

CClarify

If decision information is missing, the route asks one specific blocking question and the project status remains intact.

SAFELY CLARIFIED
AApproved

After human approval, the allowlist writes exactly one Decision, one Risk and one Milestone, including audit receipt.

3 PERMITTED RECORDS
BStale

A context change makes the previous approval demonstrably stable; the revision check keeps all project records intact.

0 STALE WRITES
RReset

The shielded operator route returns the approval state to a clean, re-inspectable starting position.

RESET PROVEN

The larger system

Connected to work

The workflow lives in a complete Business OS environment where commercial context, projects, customer information and decision records come together.

Stradviso Business OS dashboard with projects, leads, opportunities and risks.
OWNER BACK OFFICEFrom pipeline to active project records.
Stradvisocustomer portal with workflows, decision moments and project roles.
CUSTOMER PORTALThe customer sees released information, roles and decision moments.
Stradvisocustomer portal with released documents and approved updates.
GOVERNED COLLABORATIONDocuments and updates continue to be released in a controlled manner.

Product domains

In Business OS

Seven connected domains bring together commercial context, project work and controlled collaboration in one working environment.

Authenticated back officeSecure owner workspace with role-based access.
CRM & pipelineLeads, opportunities and follow-up in one commercial context.
Project workspacesActions, decisions, risks and milestones per project.
Customer portalReleased information, roles and decision points for customers.
Documents & collaborationControlled released files and project updates.
NL / EN shellThe same product logic in a Dutch and English interface.
Audit & release controlsReproducible changes, receipts and release checks.

Before, during and after Build Week

Build quickly without blurring the provenance

Build Week accelerated an existingStradvisodirection, delivered a scoped release, and now provides a frozen reference for further development.

FOR

Strategic starting point

Name, consultancy concept, limited go-to-market materials, an earlier growth scan and the first visual direction already existed. The product structure and service architecture were developed during Build Week.

DURING

Working product release

Business OS, owner back office, customer portal, Decision Studio, approval flow, tests, dual database CI, release proof, Devpost submission and demo.

AFTER

R30 remains frozen

The submitted release remains isolated and unchanged. Further developing the publicStradvisosite, live AI lane and commercial validation.

Hardest choices

Limit, recheck, prove

  • Determine what AI should never perform independently.
  • Binding approval to one specific project revision.
  • Point product, tests, evidence, video and jury access to the same release.

Lessons learned

Speed requires explicit direction

  • First build one complete vertical route.
  • Grow evidence and product behavior simultaneously.
  • Reserve capacity early for UAT, video and feature freeze.
  • Human authority belongs in architecture, not in a disclaimer.

Human + Codex + model

Intensive collaboration, human decision-making

I bore final responsibility for content and product: from customer problem and user route to governance boundary, UAT and the final submission. The collaboration did not stop with the submission; the same design discipline guides further development.

Stefan

Problem definition, product vision, user route, visual identity, human authority, UAT, voice-over and final product and submission decisions.

CODEX

Technical construction partner for architecture, interfaces, workflows, tests, adversarial checks, documentation, error analysis and release preparation.

GPT-5.6 SOL

Processes authorized context, requests targeted clarification and prepares three bounded routes; never approval or project changes.

From brand to working system

One identity, two clear functions

The publicStradvisosite makes the advice promise understandable. Business OS brings that promise to daily execution and controlled project decisions.

Stradvisowebsite spatially presented on desktop, tablet and mobile.
PUBLIC BRAND EXPERIENCEFrom a complex growth question to a testable route.
VisitStradviso
EarlyStradvisopositioning outline with an initial direction for market approach and service provision.
FOUNDATION / POSITIONINGAn early positioning direction formed the starting point for the later service architecture.
Stradvisointeractive growth scan with an initial question about the phase of an organization.
FOUNDATION / GROWTH SCANAn earlier interactive route already made the advice logic tangible.

Technical evidence note

Public evidence

The entire repository environment was shared privately during the official review.

01
Product presentationShows the core route directly in this case
 02
Devpost submissionPublic product context and full Build Week entry
03
Release and governance evidenceR30 · 473 tests · 6,431 assertions · SQLite + MySQL · 3 allowed records · 0 stale writes
Open the technical evidence note

Architecture. Authorized context → clarify of three options → human approval → revision check → atomic allowlist → audit receipt.

Test evidence. The R30 release recorded 473 tests, 6,431 assertions, and successful CI runs against SQLite and MySQL.

Evidence base. The results shown are traceable to the frozen R30 release, the recorded test evidence and the official submission.

Why this is business relevant

Human control remains visible

Decision Studio is designed for requests with timing, budget, capacity, or risk implications. It does not automatically prevent every wrong decision, but makes visible which context has been used, who has decided, what exactly is allowed and whether the approval still belongs to the current project status.

Strategy × AI × Execution

Strategy to execution

Discuss a projectBuild Week entry View more cases

Next cases

From controlled AI to other working systems

See how product thinking, technical execution and clear evidence are reflected in other projects.

AURELIA 67anchored in clear tropical water. Digital twin / luxury experience AURELIA 67 A private digital twin that connects form, materials and ownership experience. Modern villa in Chiang Mai with swimming pool and tropical garden. Thailand / property intelligence DOMORA A quadrilingual real estate toolset for the Chiang Mai market that connects listings, local housing requirements, map search, matching, seller intake and social content. BaitFeverlive on a desktop monitor, tablet and smartphone. Conceptcase / merk & product BaitFever Live brand and commerce experience for anglers; product lines, storytelling and the commercial market route are further expanded in a controlled manner. AVENLOvoice agent interface with four agents, call transcript, and bounded business actions. Voice agents / business operations AVENLO Four real-time voice agents, four brand worlds, and one bilingual phone experience with bounded business actions.
Screen example
EnlargedStradvisoproduct interface.