# Kanopi Lens

- **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:** [Ahmad Hadi Setiawan](https://cerebralvalley.ai/u/hadisetiawan)
- **GitHub:** https://github.com/ahmadhadist/kanopilens
- **Demo video:** https://youtu.be/tIALRPP9V7M
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/106

Rare aroids (variegated Monstera, Philodendron, Anthurium) sell worldwide for hundreds to thousands of dollars per plant, yet buyers judge them from a few seller photos. Variegation is graded by eye. Photos can be cherry-picked or not even show the plant being sold. Damage in transit turns into he-said-she-said disputes. As a grower in Indonesia building Kanopi, a marketplace for local and export plant buyers, I deal with these problems firsthand.

Kanopi Lens is a trust layer for trading living plants:

• Objective grading: SAM 3.1 isolates every leaf. Variegation is then measured per leaf from the pixels inside each mask. The result is deterministic and reproducible, not a guess.
• Proof of Plant: the seller records a 10–15 second walk-around video. SAM 3.1 tracks every leaf across frames, so the grade covers the whole plant rather than its best angle. The SHA-256 of the original recording is stored as evidence.
• Health scan: Muse Spark inspects each labelled leaf for pests, disease, deficiencies and stress, and returns structured findings with confidence and concrete care actions.
• Arrival Check: the buyer scans the plant on arrival. Kanopi Lens compares it with the seller's scan (leaf count, variegation, health) and asks Muse Spark whether it is the same plant and what shipping damage is visible. The result is a verdict backed by evidence.
• Buyer care guide: Muse Spark writes care instructions in the buyer's language, adapted to the climate of their country (10 languages, 16 export markets).
• Grading Card: a public link with rich previews that sellers attach to a listing and buyers can trust.

It is production-ready and API-first: FastAPI on Google Cloud Run, Firestore, Cloud Storage, API keys, per-IP rate limits, structured logging and retries. Kanopi is the first client, and nurseries, exporters and other marketplaces can integrate the same API.

I tested it on 21 photos and 9 walk-around videos (including 4K) from our nursery. That surfaced real-world problems that sample photos never showed, and each is now handled: leaves of background plants (a second SAM pass finds the potted plant being graded), red-leaved plants read as pink variegation (pink only counts for pink cultivars), 4K videos above Cloud Run's 32 MB request limit (direct upload to Cloud Storage), and photos of the same plant from a new angle (Muse confirms identity before a small variegation shift is flagged).

Live demo: https://lens.ahmadhadisetiawan.com

## More from Meta Global AI Developer Hackathon

- [Anomer](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/103)
- [KodVani](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/104)
- [Joestar](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/105)
- [CO](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/107)
- [NeuroLab AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/108)
- [Old Forge Technologies](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/109)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/106. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt