Submission for Hacktoberfest Week 2 Side Hustle. FlipPrice turns a photo plus a few fields into a listing draft, a fair-price range from sold comps, and net payout after fees for vintage resellers. Open models, zero per-listing cost. Repo MIT: https://github.com/firfircelik/flipprice

What I Built. Photograph the find, enter name brand category condition age and buy cost. Optionally upload sold-history CSV so estimates learn your niche. Get a listing draft from a local open-weight model, a fair-price range, and net payout, plus a profit dashboard. Who pays: resellers listing 20 to 50 items weekly, saving 10 minutes per listing.

Demo. Tested on Apple M2, Python 3.14. Clone repo, venv, pip install, ollama pull gemma3:1b, uvicorn app.main:app. Deploy to Render button in README. Verified live: Gemma wrote Canon AE-1 listing, p50 184.62, net 110.62, margin 59.9 pct on a 50 buy.

Code. https://github.com/firfircelik/flipprice. FastAPI plus single-page UI with photo upload, inventory, dashboard. Ollama inference, TabPFN pricing with Ridge fallback, local JSON or Atlas. Tests 7 of 7 green as user journeys: market flow, offline mode, hostile inputs, photo attach, live model test.

How I Built It. Describer: Gemma 3 via Ollama, constrained prompt, brand-first titles, local fallback chain. Pricing: TabPFN first, Ridge fallback after RandomForest failed monotonicity on 20-row CSVs. Sellers upload own CSV with validation. Note: TabPFN 9 needs one-time license plus token, then same code switches engines.

Why Open Matters. Closed APIs at 5 to 15 cents per listing erase margin at 50 listings a week. Local open weights cost near zero, work offline, keep data on device. TabPFN learns your own history. Open makes the unit economics work.

Agent Session. Built with agent assistance. ## First User. No external user yet, so full honesty: I dogfooded it as a simulated seller with 3 realistic finds, all appraised live through Gemma 3. Canon AE-1 film camera, 60 buy, p50 286.34, net 189.12. Levis 501 like new, 15 buy, p50 108.85, net 79.70. Sony WH-1000XM3, 40 buy, p50 116.82, net 61.63. Every listing got a clean brand-first title, every margin positive. Full log: FIRST_USER_TEST.md in the repo. A real reseller test is the next step. ## Prize Categories. Gemma featured, verified live. TabPFN featured, token-gated. Atlas partner. Render featured.