# Altovex CargoLinx

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Allan Sande](https://cerebralvalley.ai/u/allansande74)
- **GitHub:** https://github.com/allansande74-design/cargolinx
- **Demo video:** https://allansande74-design.github.io/cargolinx/demo/
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1526

CargoLinx is an offline-first freight desk for international trade into, out of and across Africa. For imports from Shanghai, Ningbo, Shenzhen, Nhava Sheva or Jebel Ali, it prices a consignment through to the African discharge port and on to the door (sourcing, sea freight, duty and levies, VAT, clearing, inland transit and last mile) as one delivered-duty-paid figure. For exports from East Africa, it prices the reverse on CIF terms: inland haulage, export clearing, sea freight and insurance to the buyer's destination port, where the importer takes over. Every price is calculated on a phone with no network, and the desk follows the consignment to delivery.THE PROBLEM
A small East African freight forwarder quotes a cross-border job by hand: duty rates from memory, freight from an old spreadsheet, a margin guessed under pressure. Quotes take hours, duty errors are found by the customer rather than the desk, and work stops when the network drops, which on the Northern Corridor happens frequently. The cost is national as well as commercial: Global Financial Integrity estimated Kenya lost $907 million in revenue in 2013 to trade misinvoicing, including $324 million in uncollected VAT and $229 million in customs duty.

WHAT WE BUILT
- Quote engine: a landed-cost ladder from dated statutory rates (duty, IDF, RDL, VAT, clearing, transport, last mile). Every line traces to a published rate and the date it was read. The price is firm, with a validity date.
- Documents: quotation, delivery note, packing list and invoice generated in the browser.
- Corridors: Mombasa-Kampala, Nairobi-Kampala, Mombasa-Nairobi and Nairobi-Kigali priced live, with distances, border days and arrival windows.
- Offline-first: the desk prices, quotes and queues work with no network, and syncs on reconnect without duplicates.

META MODELS, LIVE
IN THE PRODUCT — live on the public desk
https://altovexcargolinx.allansande74.workers.dev

1. Muse Spark 1.3: explains each quote in plain language and reads free text into shipment fields.
   ▶ TRY IT LIVE: open a quote and ask amanda in the chat bot "Explain this quote". You get a plain-words breakdown of the landed cost, duty and VAT.
   ▶ TRY IT LIVE: type a driver's note (e.g. "12 pallets dry cartons, ICD Nairobi to Kampala warehouse") and chat her up in text or voice "Read the note into fields". Origin, destination, cargo and quantity fill in.
   ▶ TRY IT LIVE: Ask Amanda, the assistant on every page, anything about a quote or a corridor (e.g. "Quote DDP Nairobi to Kampala for 27 tonnes of rice"). She answers with figures computed by the engine, never invented ones.

2. Muse Voice Transcribe: turns a driver's voice note into text, so a shipment is logged without typing. The desk resamples a phone recording to 16 kHz mono in the browser first.
   ▶ TRY IT LIVE: press the voice button and speak a short note (needs a microphone).

3. SAM 3.1: segments and counts cargo in a photograph (e.g. 15 pallets in 2.3 seconds).
   ▶ TRY IT LIVE: attach a cargo photo and press "Count cargo from photo".

4. Muse Wonder: a blind judge. Three drafts of an answer are written in parallel, and Muse Wonder picks the best without knowing which prompt produced it. The engine owns every figure, so the judge can change the wording of a quote but never its number.
   ▶ TRY IT LIVE: open the Intelligence screen and run "Three drafts at once, one judge". The winner is marked "judge's pick".

5. Muse Image: draws an illustration of the load, so a customer can see what is being shipped.
   ▶ TRY IT LIVE: on the Intelligence screen, press the draw-the-load button.

The desk's own proxy reaches nine Meta models; the two contributor models are refused at the proxy, because client data never goes to them.

HOW WE USE THEM SAFELY
The models never set a price. Every figure comes from our pricing engine and a dated statutory rate; Muse Spark is given those figures as the only numbers it may use. The Meta key lives in a Cloud flare Worker secret, never in the browser. The proxy admits an allowlist of models, refuses the contributor models, caps output tokens and rate-limits each address. Live calls on 30 October and 3–4 November are recorded with dates, timings and HTTP status in docs/evidence/ in the public repository:
https://github.com/allansande74-design/cargolinx

IN THE BUILD
Cargolinx was built by the Founder and CTO Mastor Sande Allan with Amanda and her agents handling 90% and cto 10% human intervention and is operated with Amanda, an AI orchestrator agent. From 31 October to 3 November Amanda ran on Muse Wonder through the Meta Model API (1,815 calls), planning features, writing and testing code and preparing documents, with every client-facing or irreversible step approved by the founder and project CTO. Key Agent roles;
- Muse Spark explains a quote in plain words and reads a driver's note into shipment fields.
- Muse Voice Transcribe turns a voice note into text.
- SAM 3.1 counts cargo in a photograph.
The models never set a price: every figure comes from the engine and a dated rate, and the model is given those figures as the only numbers it may use. The Meta key lives in a server-side secret, never in the browser; contributor models are refused at the proxy.

BUILT BY AN AGENT
Cargolinx was built and is operated with Amanda, an AI orchestrator agent working under the founder's approval for every client-facing, financial or irreversible step.

PILOTS
Four East African buyers have committed in writing to run shipments through the desk, including Brian Ojongo, Contracts Manager at Milicon's Ltd (sports equipment, China to Nairobi), and Tumwesigye Michael, Managing Director of 2Michaelz, Uganda. Grace Muhwezi of Urban edge Landscaping Nairobi, Kenya Others are described by role only until they approve public naming.

HONEST LIMITS
Four inland corridors priced; statutory rates are dated snapshots needing monthly refresh; quotes in KES only; exports not yet modelled; one operator per browser for now and exports are priced CIF to the destination port; destination duties and delivery are the importer's.

Live desk: https://altovexcargolinx.allansande74.workers.dev
Code (MIT): https://github.com/allansande74-design/cargolinx

## More from Meta Global AI Developer Hackathon

- [meta_hackathon_winners](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1523)
- [ProbeForge](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1524)
- [Sentra-X](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1525)
- [Keep Talking](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1527)
- [Neural Referee](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1528)
- [OjO](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1529)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1526. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt