Photon 2.0 hackathon track: the judging rubric and eligibility rule, and how a utility-style iMessage bot can answer them
From Photon's 'Hackathon Guidelines for Photon 2.0' page (Notion; an automated fetch returned an empty shell, so a teammate opened it and pasted the text, Nov 2026). Eligibility: 'your project must be built on Photon 2.0 and live on at least one Photon channel'; iMessage is the flagship text channel, voice agents run on an SMS number (calls and group chat on iMessage need a dedicated line). Judging: Vision (hybrid intelligence), Craft (cohesive, responsive, tastefully integrated), Depth (real-world context and edge cases), Traction (can we imagine real users tomorrow). 'What great looks like': real human-in-the-loop collaboration rather than a demo reel of prompts; context-awareness that persists across time, channels and teammates; a delightful default that elevates the moment without demanding attention; thoughtful safety and fallback behaviours that earn trust. Prizes: iMessage winner, runner-up, and a separate Best Voice Agent prize. Note the page's SDK link points at spectrum-ts while the 2.0 beta Python package is photonhq-api. If your bot is a lookup tool rather than a companion, map concrete behaviours to their wording in the write-up: confirmation before acting on a photo ('Is that right? Reply YES') and numbered choices instead of guessing = human in the loop; remembering the current subject so follow-up commands need no name, plus saved items that trigger a later alert = context that persists over time; an instant acknowledgement before a slow step = delightful default; asking the user to type when an image is unreadable, a STOP that deletes stored data, and idempotency keys so retries never double-send = safety and fallbacks; measured reply time = responsive; one concrete first user = traction.