BriefStruct
An independent capability project using fictional studios and briefs, not client work. Explore Sample presents a stable pre-generated, read-only case. Try your own space creates an isolated 24-hour workspace with real DeepSeek analysis, human review and export. AI drafts require a person to check scope and make decisions. The screenshots below come from the live public V3 and its saved real DeepSeek result. No emails are sent.
01
Understand
Turn a messy intake into a reviewable specification.
Studio briefs arrive as fragmented notes with missing content, unclear responsibilities and undecided scope. I built an AI-assisted review workflow that converts the source brief into structured scope, questions, risks and assumptions, while keeping the original material close enough to review.
02
Trust
Check the source before accepting the AI draft.
DeepSeek returns a structured document that the server validates before saving. EXP identifies quotes verified against the saved source, INF identifies interpretations to review, and GAP exposes missing decisions. Selecting a result reveals its evidence and context. Human edits create new revisions while preserving the original AI output.
03
Act
Review, revise and take the work forward.
A visitor can explore the read-only sample or create a private space, bring in a brief, review the result and export the current human revision as Markdown or PDF. This workflow demonstrates the kind of AI-assisted internal tools, intake workflows, brief and scoping systems, and studio productivity tools I can build.
04 / Visual screens
a closer look.
Captured directly from the public V3. The saved real DeepSeek result is shown after a human revision; selecting the scope item highlights its exact quote in the original brief.
The phone interface separates Result, Source brief and Evidence into focused views. This live capture shows the checked source quote for the same saved real analysis; capturing it did not call AI again.
05 / Build notes
behind
the pages.
- Real DeepSeek integration is embedded in a brief review workflow. Model output is treated as untrusted: structured schema, enums, size limits and exact source quotes are checked before a successful result is shown.
- The review interface connects source material, structured results and evidence. EXP, INF and GAP retain their distinct meanings; human review and revision history keep decisions visible.
- Markdown and PDF are generated from the current saved revision. Download access is checked against the visitor’s own space.
- Private spaces isolate briefs, analyses and exports. Persistent quotas and fixed 24-hour expiry survive logout and restart; a separate cleanup process removes expired demo resources every 15 minutes.
- A shared monthly AI budget and per-space analysis limits control real provider dispatch. There are no automatic paid retries or hidden mock responses after a provider failure.
Explore a read-only fictional sample, or start your own private 24-hour demo space. Real AI is limited to three analyses per space and the application budget. Review the draft before exporting.
View demo ↗