AI · decision systems / Source-inspected prototype / portfolio diagram
DecisionOps
An AI recommendation can sound settled before a team has reviewed it. The source-inspected DecisionOps prototype turns meeting transcripts into proposed changes that people can approve, reject or defer.
Human judgement / in the loop
A summary
is not
a decision.
What do we know?
What are we assuming?
Compare alternatives.
Confirm the rationale.
not automatic.
Actual scripted portfolio interaction. Approve, reject or defer a proposal. No live AI, production integration or external action.
Inspect the review logic ↗Recording unavailable. Use the working demonstration linked above.
- Intended users
- Intended users: product teams and founders who need to review proposed changes against meeting evidence. Supplied research does not establish adoption by this audience.
- My responsibility
- The supplied résumé attributes transcript-to-change-set workflows, review states, exports and versioned prompts. A separate DecisionOps source project is now available for inspection; collaborator credits and dated ownership records remain to be confirmed.
- Context & dates
- Source-inspected prototype with a deterministic fictional-data demo. The supplied Distill project identifies itself as DecisionOps. Release dates and hosted operational status remain unverified.
Implementation boundary
What is actually here.
The source contains a deterministic demo with review, evidence jumps, edits and export links. It also includes transcript parsing, structured validation, an OpenAI provider adapter and workspace-scoped API code. This portfolio interaction remains an authored diagram; hosted end-to-end execution has not been verified.
02 / A reviewable change
A proposal is not
a decision.
Follow one authored change through a human review. The interesting part is what must stay visible when the wording becomes an action.
Interactive workflow diagram · authored example · no AI, meeting ingestion or connected tools
- 01Source preserved
- 02Proposal authored
- 03Human review
- 04Handoff withheld
Make review explicit.
- Authored proposal
- Explore a clear review step before handoff.
- Not yet known
- Who owns it, what success means, and whether this is the right fix.
Awaiting your review
Approval permits a proposed handoff; rejection stops it; deferral keeps the missing information open. No action here changes an external system.
A source sentence can travel through this diagram. It cannot create an external task.
DecisionOps / labeled workflow diagram
- 01Discussion
- 02Options
- 03Owner
- 04Decision
The source contains a deterministic demo with review, evidence jumps, edits and export links. It also includes transcript parsing, structured validation, an OpenAI provider adapter and workspace-scoped API code. This portfolio interaction remains an authored diagram; hosted end-to-end execution has not been verified.
The choice
and its cost.
More deliberate input; less risk of implying agreement.
What makes a discussion a decision?
An AI recommendation can sound settled before a team has reviewed it. The source-inspected DecisionOps prototype turns meeting transcripts into proposed changes that people can approve, reject or defer.
- Alternatives
- Transcript archive / Automatic decision extraction / Human-confirmed decision record
- Assumption
- Explicit ownership and review conditions may make decisions easier to revisit.
No interviews, meeting logs or usage study have been supplied for DecisionOps. The tension comes from the creative brief.
- Source
- User-supplied product direction only. This decision record is a proposed exploration.
- Constraint
- A transcript is not evidence that a team has agreed.
Proposed: a human confirms the rationale, owner and review condition.
- Rejected direction
- Treating an AI summary as proof that a team agreed.
- Trade-off
- More deliberate input; less risk of implying agreement.
The proposed interface would move from a meeting archive to a human-confirmed decision record.
- Still untested
- Whether this interaction changes behaviour. The visual sequence is a concept demonstration, not a measured outcome.
An AI recommendation can sound settled before a team has reviewed it. The source-inspected DecisionOps prototype turns meeting transcripts into proposed changes that people can approve, reject or defer.
The proposed interface would move from a meeting archive to a human-confirmed decision record.
What the evidence
can support.
Source and test definitions were inspected. This review did not call a model, apply migrations or establish customer value. Synthetic arithmetic tests and authored fixtures are not measured model accuracy. Older project documents conflict with parts of the current source; hosted behavior needs separate verification.
Available here
- Inspected app/PortfolioDemo.tsx: fictional fixtures, evidence navigation and human review states.
- Inspected lib/transcript.ts, lib/ai-contracts.ts, lib/providers/openai.ts and analysis/review/export API source.
- Test definitions for parsing, structural validation and export boundaries; definitions are not fresh passing results.
Repository paths and code boundaries · inspected 27 October 2026 · source inspection is not deployment verification.
Not yet evidenced
- Permissioned meeting examples and original discovery records.
- Hosted end-to-end execution, identity/storage configuration and isolation checks.
- Dated ownership and collaborator records.
- Real model evaluations, participant agreement and adoption.
Discuss the work.
For a role or a related product problem, start with the contribution you need.