Inspiration

Small importers, wholesalers and family traders manage scattered purchase requests, inventories and quotations. Manual comparison across folders, languages, units and specifications wastes time and risks missed or unsuitable offers.

What it does

Dubai Souk AI privately matches authorized buyer requests with supplier quotations stored on UGREEN AI NAS and returns evidence-linked trade opportunities without contacting counterparties or executing transactions.

The trader selects buyer and supplier files or folders, adds optional filters and requests matching. The Skill returns a cited shortlist classified as Exact-compatible, Conditional / review needed, Unresolved or Not compatible, and the trader decides independently whether to pursue an offer.

  • Works across Arabic/English mixed documents and keeps original-language excerpts.
  • Keeps uncertain terms visible instead of guessing (no unit or currency conversion without verified support).
  • Separates the confidential multi-supplier comparison from a draft counterparty view that excludes other traders' information and is never sent automatically.

How we built it

A Uliya orchestration Skill built with the official Uliya Skill Kit v0.7.3:

  • ug-filesystem (ls/find) through file-agent to discover documents inside user-selected scopes;
  • ug-document (extract/reconcile/qa) through document-agent to extract and compare terms, with comparison QA as fallback;
  • optional ug-knowledge (save-text) through knowledge-agent, only for a user-requested, approved report saved to an explicitly specified folder.

The standard workflow is read-only. No scripts, credentials, external APIs, transfers or trading actions. AI tools used: OpenAI ChatGPT (workflow design, documentation drafting) and Anthropic Claude (technical review against the official kit, revisions, static checks).

Challenges we ran into

Keeping every step mapped to a capability that actually exists in the catalog, handling mixed Arabic/English quotations without translating them, and preventing one trader's prices from leaking into a summary meant for another.

Accomplishments that we're proud of

The package passes the official Uliya static checker against catalog v2.0.0 with zero errors and zero warnings, and includes five fictional validation scenarios, including Arabic and privacy cases.

What's next

If selected as a finalist: install and validate on a real UGREEN AI NAS, including Arabic extraction, permission enforcement and partial coverage. Current status: static-pass only; not yet tested on a UGREEN NAS.

Built With

  • chatgpt
  • claude
  • markdown
  • ugos
  • uliya-skill-kit
Share this project:

Updates

Submission history