Inspiration
I was babysitting my godson while his parents planned a family Disney cruise. They wanted him to hold up his side of the bargain by finishing his chores. After each chore, I sent them a photo. When they came to pick him up, they told me saving for the cruise was a huge part of the goal. Saving for the whole family was expensive, but setting aside some money every month made it achievable.
That is Do My Chore. The parent names a real family goal. Nebius Token Factory with NVIDIA Nemotron turns that goal into a chore plan and a monthly save. The Token Factory photo check replaces those manual photo texts I used to send, so the kid can show the chore is done and the parent still decides.
What it does
Do My Chore helps a family turn a real money goal into weekly habits. The parent names the goal (for example a Disney cruise), the target amount, and the date. Suggest plan on Nebius Token Factory with NVIDIA Nemotron drafts a monthly save schedule plus weighted kid chores that add up to 100%. The kid sees percent progress only, never dollars. When a chore needs a photo, the kid submits a picture; Token Factory runs a photo check with Qwen3.8-27B and returns an advisory approve or reject. The parent still makes the final call, and approved check-ins move the kid percent bar.
How we built it
Flutter and Dart for the parent and kid UI, with an in-app role switch on one seeded demo family. Supabase handles auth, Postgres with RLS on every family-scoped table, photo storage, and Edge Functions. Live AI goes through Nebius Token Factory: suggest-plan calls nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B for the goal-first chore and save plan; photo-assist calls Qwen/Qwen3.8-27B for the kid photo check (this Token Factory account has no Nemotron vision model). Parent approval stays in the app. Secrets stay on the hosted project; DEMO_WALK only stubs the camera JPEG for simulator demos. Public code: https://github.com/pamu512/do-my-chore-nemotron
Challenges and learnings
Token Factory on this account returned 404 for Nemotron vision (nemotron-3-nano-omni), so photo check could not stay on Nemotron. We kept Suggest plan on Nemotron Nano 30B and moved photo check to Qwen3.8-27B on the same Token Factory path, and said that clearly in the demo and README. Edge timeouts also had to rise for live calls. The harder product lesson was trust: AI can draft the plan and advise on photos, but a parent still has to approve before any kid progress counts.
Log in or sign up for Devpost to join the conversation.