A single-commit hackathon MVP that visualizes a deterministic, agent-styled seller workflow from voice or text input through CRM, illustrative attainment, recognition, and simulated notification stages.
The workflow before the build
A spoken seller update may need to be translated into structured CRM intent, an opportunity action, an illustrative attainment view, recognition language, and a communication payload across separate steps.
What the public project delivers
The public MVP implements a sequential TypeScript pipeline with keyword-based intent extraction, a mock or configured Dataverse opportunity stage, illustrative commission and attainment calculation, recognition copy, simulated notification output, Express Server-Sent Events, and browser speech or text input.
- Accepts browser Web Speech or text input and streams stage progress with Server-Sent Events.
- Uses deterministic keyword extraction rather than an external AI model for deal intent.
- Runs a mock path or an optional configured Dataverse opportunity update.
- Produces illustrative attainment, recognition, and simulated notification payloads.
The value case, stated honestly
The prototype is designed to make a multi-stage seller workflow visible as one narrated sequence, not to establish autonomous AI behavior, production safety, or measured seller impact.
Evidence label: Designed value: a tangible visualization of how one seller update could pass through structured CRM, calculation, recognition, and notification stages.
Architecture in one view
Voice or text input moves through deterministic TypeScript stages, a mock or configured CRM and calculation step, and a simulated notification.
The browser captures a seller update through Web Speech or typed input.
Sequential TypeScript stages extract intent and compose the workflow.
A mock or configured Dataverse action is followed by illustrative attainment logic.
Recognition and notification payloads are composed, but the Teams-style send is simulated.
Human control and guardrails
- The journal labels the stages deterministic and agent-styled rather than claiming five autonomous AI agents.
- Attainment and commission values are illustrative rather than financial evidence.
- A real Dataverse deployment should add an explicit approval gate before any mutation; the current optional write path does not provide one.
Where the pattern stops
- When Dataverse is configured, speech completion starts the pipeline and the opportunity stage can write without an explicit user approval gate.
- The notification endpoint logs intent and returns success without sending a real Teams message, while the payload marks the channel simulated.
- The server enables unrestricted CORS with no authentication middleware, and browser what-if values are inserted through innerHTML without escaping.
- Date handling is hard-coded around 2026, and the repository has no LICENSE file even though package.json says MIT.
Explore the public proof
- Deal to Glory public repository Public orchestrator, voice-intake, opportunity, notification, server, browser, package metadata, README, and single-commit history support this entry.
- GitHub public repository metadata Repository created 2026-02-12; last public push 2026-02-12. These are repository facts, not adoption evidence.
- Attribution: Express Upstream web-server dependency; not owned by this repository. License: MIT in package metadata.
- Attribution: @azure/identity Upstream Azure identity dependency used by the optional Dataverse path. License: MIT in package metadata.
- Attribution: TypeScript Upstream language and build dependency; not owned by this repository. License: Apache-2.0 in package metadata.