Facts and interpretation stay distinct
Counts and due-date rules are computed locally. Generated priorities are advice for the founder to review.
PROJECT 001 / BUSINESS OPERATIONS
Mission
Control.
Turning scattered business records into a briefing a founder can inspect, challenge and act on.
01 / THE PROBLEM
A useful briefing needs
a dependable record behind it.
Beni’s company knowledge, work register, lead register and decision log were held in separate Markdown and Excel files. A conversational summary could help, but a usable operating workflow also needed persistent records, visible source gaps and a way to preserve what had been reviewed.
The practical question was simple: what needs the founder’s attention, and what evidence supports that recommendation?
The broader objective is a working portfolio that supports enterprise AI implementation, consulting, fixed-term projects and employment conversations. This independent prototype is not an enterprise client deployment.
02 / THE IMPLEMENTATION
One workflow,
with deliberate boundaries.
A local dashboard reads the four business records into SQLite. Operational changes are kept separately from the originals. Rules compute counts and dated exceptions; a manual Codex briefing provides interpretation and recommendations.
The founder exports a source pack, initiates the chat, reviews the result and records the reviewed briefing with notes. The application does not operate Codex or upload the pack automatically. The optional API adapter remains in the code, with funding deferred.
Founder-led, AI-assisted delivery
The founder set the direction, operating priorities and review requirements. AI coding assistance supported implementation, tests, investigation and documentation. Business decisions and acceptance remain with the founder.
- Operational records: tasks, enquiries, decisions and imported source snapshots.
- Traceability: source fingerprints, exported inputs, saved runs and review notes.
- Human decisions: recommendations do not send messages, publish content or complete work.
03 / ARCHITECTURE
A small system.
Inspectable connections.
Company knowledgeWork registerLead registerDecision log
↓
Mission ControlPython application · deterministic rules · SQLite
↓
Exported source packSnapshot and prompt
→Manual Codex chatFounder initiates analysis
↓
Founder reviewCheck evidence · correct advice · record notes
↓
Saved reviewed briefingManual provenance retained · API status unchanged
Current manual path. The optional API path is separate and has not produced a verified live briefing.
Python standard librarySQLiteHTML / CSS / JavaScriptCodex manual workflow
04 / CONTROLS
Accountability is part
of the workflow.
A result keeps its source context
Exports retain the input snapshot. A returned manual briefing references the export used, with reviewer and review notes.
Stale exports are rejected as current
If records change after export, the manual intake requires a fresh pack before the result can be saved as current.
Review is not automatic verification
Pasted manual results are not independently checked for factual support by the application. The API adapter has separate source-ID, quotation and task-selection checks.
No consequential action tools
The briefing workflow cannot contact prospects, publish material, spend money or mark recommendations as executed.
05 / EVIDENCE AND LIMITATIONS
Show what passed.
Keep the limits visible.
32
Tests passed in the inspected local baseline
Import, persistence, isolation, provider diagnostics and the manual export/review workflow. JavaScript syntax checks also passed.
Latest reconciliation update
The founder’s transferred working copy reports 41 passing tests, including nine new checks for approved strategy and priority reconciliation. That updated evidence has not been independently inspected in this website build. Task acceptance remains under founder review.
| Evidence | What it establishes | Boundary |
|---|---|---|
| Automated checks | Selected local behaviour and mocked provider responses | Not live model quality or production resilience |
| Manual briefing pack | Source-linked handover and a reviewable operating process | Founder-initiated, not autonomous execution |
| API failure diagnostics | A live credit-balance failure was identified specifically | No accepted live API briefing; funding deferred |
| Release documentation | Requirements, decisions, runbooks and release gates are recorded | Documentation is not evidence that a gate has passed |
No quantified productivity gain, client outcome, revenue result, uptime or enterprise deployment is claimed. The public walkthrough uses fictional records and prewritten outputs.
Download the public case study06 / RELEASE READINESS
The current decision:
continue the prototype.
Mission Control is a local implementation under development. The current production decision is no-go. The public AI Lab is a static portfolio; it does not expose the operational application or its private database.
Local operational workflowImplemented
Reconciled strategy and prioritiesReported; evidence review pending
Task acceptance criteriaFounder review in progress
Live API briefingDeferred
Production security and recoveryOutstanding
Before a private enterprise pilot: establish the access model, managed secrets, approved data handling, controlled deployment, tested recovery, monitoring and measurable quality/cost acceptance.
07 / LESSONS AND NEXT STEPS
The engineering is
in the decisions.
Integration readiness comes before model quality.
Initial live attempts exposed network restrictions, then a credit-balance failure. More precise diagnostics replaced an ambiguous quota message with an actionable cause.
A manual workflow can make the process testable.
Deferring API funding does not prevent evaluating whether a briefing is useful. Manual handover makes the source pack, founder review and operating value visible before introducing automation.
Updated decisions must reach the application.
The reconciliation work addresses obsolete strategy text and priority ordering. This is a reminder that an approved business decision and its implemented system behaviour require separate evidence.
Inspect the process yourself.
Explore the fictional walkthrough or start an implementation conversation through Beni.