Introduction

Kaya AI is built around Amber, Kaya's procurement agent. Amber learns project context and works across supply chain, procurement and last-mile delivery in construction. Amber coordinates the work: each specialist agent does one job very well, and Amber hands work to it. Every Grand Challenge entry is built so Amber could rely on it.

Tools and Technologies

  • Round 1: use your own tools, any model, and public or self-prepared data. Kaya mentors hold office hours: [schedule]
  • Round 2: details are shared with the top 10 teams when they are announced
  • Discord, for announcements, team formation and questions: https://discord.gg/q9xtXfKS

Inspiration

  • How could one agent take a job it has never seen, work out the steps, and know exactly when to stop and ask?
  • How could Amber split a big job across specialist agents, and catch it when two of them disagree?
  • How could an agent keep a plan on track when a vendor slips a date, a price changes, or a tool call fails?
  • How could a site walk recording turn into updated delivery dates and action items, with nobody typing?
  • How could a photo of a delivered generator be matched to its order line, its laydown area and the work that needs it?
  • How could a worker on site point a phone at a piece of equipment and know straight away whether it belongs in that room?
  • How could an agent compare eight bids with different scopes and exclusions, and cite every number?
  • How long will a transformer really take to arrive, and how sure can you be?

A Note on Judging

The Grand Challenge has two rounds. Both divisions are scored the same way and ranked separately.

Round 1 (Build, 30 days): The Kaya team, together with Edition 1's winning teams, tries each agent on realistic examples for its challenge. The top 10 teams go to Round 2.

Criterion Weight What judges look for
Accuracy 40% Does it get the job right on realistic examples?
Customer value 30% Would a procurement manager or superintendent use it tomorrow, as a product?
Trustworthiness 20% Sources cited, sensible confidence scores, nothing made up
Build quality 10% Easy to run, cost and speed per run, code quality

An agent that says "not sure, check with a human" scores better than one that confidently gives a wrong answer.

Round 2 (Finals): The top 10 teams take their agents further, working closer to how Amber works. At the finale in Mumbai on December 17, each team demos its agent live before Kaya's CEO, Ojonimi Bako, and our panel of judges (to be announced). Finalists are judged on the same criteria, with extra weight on how much the agent improved during Round 2.

Contact and Support