Demonstration project — fictional brand

MUNGO Fret & Transit

The website of a parcel forwarder, with an AI assistant that answers on a Sunday night and knows when to hand over.

Parcel shippingParis, Brussels and DoualaEurope and Cameroon

Home page of the MUNGO site on desktop, in French, with a field to ask the assistant a question and a parcel label
The same home page on mobile

Project sheetMUNGO

Sector
Parcel shipping
Market
Diaspora and Cameroon
Type
Website and AI agent
Pages
14
Languages
French, English
Built in
2026

The need

MUNGO ships families’ parcels between the Paris region, Brussels, Douala and Yaoundé. Its advisors spend the day answering the same questions: how much, when, is it allowed? And families often pack their boxes in the evening or on Sunday, when nobody is there to answer.

For their customers

On a Sunday at 11 pm, Awa is packing two boxes for her mother in Douala. She asks the assistant for a price and gets it straight away, air and sea side by side. If her parcel arrives damaged, the assistant puts her in touch with a person, with a summary she can read first.

What I built

The agent is simulated for this demo. In production, a language model would use the same tools and guardrails, with human oversight.

  • The assistant compares air and sea for two 25 kg boxes; on the right, the panel shows the quote tool it called

    An assistant that asks before it calculates

    For “2 boxes of 25 kg”, it first asks for their size, because volume changes the price. Then it compares: €475.00 by air, €144.00 by sea.

  • Answer about perfume, prohibited by air and accepted by sea; the panel shows the corrected typo, the intent and its confidence

    Its reasoning, next to the conversation

    The “How the agent reasons” panel shows every turn: the typo it fixed (“perfum” becomes “perfume”), what it understood, the tool it called and its source.

  • Hand-over to an advisor: the editable summary of the conversation, before choosing a channel

    A person, when it matters

    A complaint, customs duties, a large volume: it hands over. The customer reads the summary, then picks WhatsApp, email or a call-back. Nothing is sent until she clicks.

  • The prices page, in French, with the rates of 1 October 2026 and the button to estimate a shipment

    The same tools as the pages

    The price estimator, parcel tracking and the departures calendar use the same functions as the assistant. It cannot do anything the site cannot do.

Design choices

The codes of freight: parcel labels, city codes, a weighing stamp. Navy blue and container orange feel as reassuring as a well-run port, and the agent’s panel reads like a logbook.

  • Red Hat Display for headings and text
  • Red Hat Mono for tracking numbers and the agent’s logbook
  • ConcreteBackground#EEF0F2
  • InkText#0E1A27
  • OrangeAction#F26A1B
  • NavyBands, conversation#0B2A4A

Measurements

Home page of the demo, on mobile and on desktop.

Lighthouse scores and size of the MUNGO home page
MetricMobileDesktop
Performance95100
Accessibility100100
Best practices100100
SEO100100
Largest Contentful Paint (LCP)2,064 ms517 ms
Transferred size201 KB297 KB

Measured on 8 October 2026 with Lighthouse 13 (default profiles), page served compressed from a local server, as in production.

Built with AI

Design, art direction and checks by me. Code and content produced in collaboration with AI agents, then reviewed and tested.

How I work with AI

Let’s talk about your project.

Tell me what you do and what you need. I reply within one business day.

You talk to me directly

Mike SOBTAMFO · engineer · Cameroon (UTC+1)