Inspiration
Most households have items that are not quite trash but are easy to overlook: a scratched chair, old books, a faded shirt, or an empty glass jar. We wanted to make the circular-economy decision simple and practical: help people see the best next life for an item before it reaches landfill.
What it does
HomeCycle AI helps users choose a sustainable next action for everyday household items: repair, reuse, donate, resell, upcycle, or recycle.
Users select an item, explore several visual recommendations, and open the option they prefer to see estimated carbon saved, landfill avoided, resale potential, and practical reasoning. A dashboard and session history make the cumulative impact visible without requiring an account.
How we built it
We built HomeCycle AI with Next.js, TypeScript, and a responsive UI designed for quick decision-making. The demo includes six curated scenarios: a wooden chair, books, laundry basket, smartphone, glass jar, and T-shirt.
Codex helped us review the codebase, improve the UI flow, build the static demo experience, integrate visual assets, refine documentation, and verify production builds.
For the production AI path, we implemented an OpenAI Responses API route using GPT-5.6. It is designed to accept an image, identify the item and condition, choose the most appropriate circular action, estimate impact, and return strict structured JSON for reliable rendering. The public demo currently uses curated deterministic results in the same structure so it remains dependable during judging.
Challenges we ran into
The biggest challenge was balancing a compelling AI experience with reliability. Image analysis depends on API availability and billing quota, which can interrupt a live demo. We solved this by preserving the GPT-5.6 production architecture while presenting curated scenarios that always work.
We also worked to keep the interface focused. Early versions showed too much information at once, so we changed the experience to reveal detail only after users select an item or recommendation.
Accomplishments that we're proud of
- Turned sustainability guidance into a clear, visual decision flow.
- Made impact understandable through carbon, landfill, and resale estimates.
- Created a polished no-login demo that works consistently.
- Built a GPT-5.6-ready structured-output architecture for future live image analysis.
- Used Codex throughout development for implementation, UX refinement, testing, assets, and documentation.
What we learned
We learned that trustworthy AI products need more than a good model prompt. They need predictable UI states, structured outputs, graceful fallbacks, and a clear explanation of what the recommendation means.
We also learned that sustainability tools become more useful when they connect values to practical choices: repair the chair, donate the books, trade in the phone, or reuse the jar.
What's next for HomeCycle AI
Next, we plan to enable live image analysis with GPT-5.6, add location-aware donation and recycling options, and connect a persistence layer for longer-term impact tracking.
We also want to add item condition checks, local resale recommendations, and personalized progress goals so HomeCycle AI can become a useful everyday companion for reducing household waste.
Log in or sign up for Devpost to join the conversation.