FlipPrice

The FlipPrice project has been open-sourced on GitHub (existentialcoder/flipprice), including an Expo client, a FastAPI service, and a demo video made with Remotion, and its competition categories cover three directions: Best Use of TabPFN, Best Use of Gemma, and SerpApi.

FlipPrice is an open-source project submitted by existentialcoder in the Hacktoberfest open-source AI challenge; users take a single photo and get a suggested price and range, the 5 closest sold records, ready-to-publish listing copy, and the tool can learn the user's pricing habits.

FlipPrice's tech stack is: Gemma 4 (vision) handles visual recognition and generating listing copy, SerpApi pulls eBay sold data, TabPFN performs personalized pricing regression locally, MongoDB stores sales history, FastAPI provides the backend interface, and Expo builds the client.

In the FlipPrice demo, the first version's pricing was 65 dollars, and after filtering and fixing, the final price was 28 dollars (range 27–39 dollars), based only on 3 truly matching eBay sold records.

FlipPrice degrades to brand matching when eBay sold data is insufficient, and only uses all search results when there are fewer than 3 matches; this filter corrected a perfume valuation from 65 dollars to 28 dollars.

The developer believes that recognition errors are more costly than pricing errors, because incorrect brand recognition leads to searching for irrelevant products and giving an incorrect price reference.

The developer believes that open source is crucial for this kind of application, because pricing depends on private data (purchase price, sale price, buyer preferences); a closed-source API means uploading historical data and getting a non-personalized market perspective, whereas open weights allow running the model locally, avoiding data leakage and per-image billing.

The author once tried RandomForest as the pricing model, but it exhibited a monotonicity failure on a small 20-row sample CSV, so it was abandoned.

FlipPrice is an open-source project under the MIT license, positioned as a "side-hustle pricing copilot," with its code hosted on GitHub (https://github.com/firfircelik/flipprice).

The Gemma 3 1B small model used by default in FlipPrice has limited generation quality, TabPFN's paid threshold is a breaking point for purely free users, and the lack of real user data makes valuation accuracy questionable.

FlipPrice's inputs include product photos as well as fields such as name, brand, category, condition, era, and purchase cost, and its output is divided into three parts: a listing draft, a fair price range, and the net amount received.

FlipPrice's listing draft is generated by a local open-source-weights model, with the title starting with the brand.

FlipPrice has been verified on Apple M2 with Python 3.14, and the installation process is to clone the repository, create a venv, pip install, ollama pull gemma3:1b, start uvicorn, and it also provides a one-click Render deployment button.

FlipPrice's description generation uses Gemma 3 (via Ollama) and has a local fallback chain to support offline operation.

FlipPrice's pricing engine uses TabPFN first, with Ridge regression as a fallback.

Gemma's hosted model API once returned a 500 error during testing; the FlipPrice backend retries and falls back to a second model, while recording which model generated each response.