Scope-change Desk

• 

  Saved local quote revision with synthetic inputs. Exact amounts are resolved; client approval remains unrecorded.

• 

  Mobile saved-draft view using synthetic quote inputs. No client message or approval action was performed.

The problem

A client asks for two more pages and a checkout after a website quote has been agreed. The change seems small in a message, but its price, delivery impact and approval are separate questions. Scope-change Desk makes those unresolved decisions visible.

The product

Start from a structured quote, paste a change request, inspect suggested changes and literal message citations, then review and save a versioned draft. Existing rates remain explicit. New work stays unpriced until a rate is supplied, and an unknown line keeps the full total unresolved. Integer minor-unit arithmetic avoids floating-point money errors. Saving a draft never implies client approval or a validated schedule.

Sponsor technology and build

NVIDIA Nemotron Lightning (nvidia/Nemotron-3_5-Lightning) on Nebius Token Factory interprets selected message text and baseline descriptions/quantities. Node.js validates the structured response and performs deterministic quote arithmetic. The user reviews suggestions before applying them.

A real Tavily search retrieved technical documentation for requested checkout work. Selected passages fed a separate Nemotron clarification checklist. Research does not infer rates, contact a client or approve an agreement. Deepgram Aura-2 Thalia generated the narration; technical audio checks passed, while listening review and the edited video remain pending.

What is verified

The actual workflow completed live message interpretation and saved a reviewed draft. The standalone full suite passed 78 tests; later focused checks passed 17 research units and two API/browser checks. Counts overlap; local browser and fixture tests do not substitute for live model evidence.

The first research output used number-only citations and failed semantic review. A later live output contains substantive quotations but overstates fulfillment guarantees and rendering behavior. Both are preserved. A separate assistant review proposes narrower source comparisons; neither model research version has a semantic-quality pass. The demo uses a synthetic freelance request and establishes no real customer approval or payment integration.

Run and current access

Repository. As checked on 22 October 2026, this repository is private and its remote revision predates the validated local release. Publication and judge access remain pending; the repository link is not yet a public reproduction path.

The commands below apply to the prepared standalone release after access to that revision is available. Node.js 22+ is required. Live sponsor actions need private environment configuration and explicitly reviewed provider access; local examples work without provider keys.

npm ci
npm start

Open http://127.0.0.1:4190/scope. For local fixture tests, run npx playwright install chromium and npm test without provider credentials. The loopback application, quote revisions and evidence are implemented; public judge access and the final demo video remain pending.

AI tools and development

Codex assisted with implementation, interface refinement, test authoring, debugging and independent review. NVIDIA Nemotron performed the bounded runtime tasks described above; Tavily supplied search results, and Deepgram generated narration. AI output was checked against saved state, validation rules and actual execution evidence. This description does not claim that another assistant's separate hackathon projects are part of this entry.

Built With

• css
• deepgram
• html
• javascript
• nebius-token-factory
• node.js
• nvidia-nemotron-lightning
• playwright
• tavily