Noon Arch AI Operations

Inspiration

Engineering and service teams lose hours chasing approvals, payments, meetings, expiring documents, and supplier quotations by phone. Those conversations are hard to audit and usually disconnected from the task system where the work began.

What it does

Noon Arch AI Operations turns a manual or ClickUp record into a reviewable CALL-E phone workflow. An operator selects one active work item and one recipient, reviews the exact objective and AI disclosure, and explicitly approves one execution. The live server route then calls CALL-E and normalizes the outcome into structured fields that can be reviewed or written back to the linked task.

The product currently includes four configurable workflows:

approval and payment follow-up meeting scheduling with attendee conflict checks employee document and license expiry reminders supplier quotation collection with confidential price negotiation and comparison

How we built it

The TypeScript/React application uses a provider boundary that keeps business-system reads and writes separate from the CALL-E execution gate:

provider discovery -> saved binding -> active-task refresh -> recipient match -> availability check -> draft import -> manual confirmation -> CALL-E -> structured result -> optional write-back

ClickUp bindings discover Workspace, Space, Folder, List, status, attendee, date, phone, and custom fields at runtime, so operational IDs are not hard-coded. Cloudflare D1 stores application settings, contacts, call records, and encrypted connector credentials.

Safety and product experience

explicit AI disclosure and one-recipient approval before every call confirmation reservation and provider idempotency protection no automatic retries or hidden recurring calls E.164 validation and masked histories optional ClickUp comment write-back meeting start/end/status updates only after an agreed slot passes a fresh conflict check configurable organization, assistant, region, locale, timezone, contacts, sources, and mappings English and Arabic interface credential-free deterministic demo with fictional records and zero external side effects

How AI and Codex were used

CALL-E conducts the approved outbound conversation at runtime. Workflow context is assembled from the selected source record rather than a fixed script, and results are normalized into workflow-specific structured outputs.

Codex helped design the connector architecture, implement CALL-E and ClickUp integration, refine the bilingual interface, diagnose provider-region behavior, build confirmation and duplicate protections, create 23 regression tests, audit the public source for secrets, prepare the upstream contribution, and produce the demo assets.

Testing

The public judge path requires no credentials or phone credit: choose a workflow, review the fictional record and masked recipient, approve the disclosure and execution, run the clearly labeled no-call replay, and inspect the structured result and write-back preview.

Local verification: npm ci npm run audit:public npm test npm run dev

All 23 tests, the production build, public-source audit, and the official upstream repository validator pass.

Built With

• call-e
• clickup-api
• cloudflare-d1
• cloudflare-workers
• drizzle-orm
• react
• typescript