Hot dog. Or not hot dog.
Send an image and a typed Choice question through the Venn Decisions API. No training or photo upload to a cloud service.
1. Ask Venn without task training
Send a prompt and the selected image to Venn; inspect the typed response.
2. Train a custom family
Read your labeled photos, then fit a linear family. The reader's weights do not change.
No fitted family yet.
3. Evaluate separately
Measure accuracy on labeled test photos never used in fitting. This does not train the family.
No evaluation yet.
4. Run repeatedly
Choose a photo and reuse the saved family. Image reading and decision scoring are timed separately.
Cached vectors are labeled explicitly. Probabilities are not an accuracy guarantee.
Your local photos
Training, test, and demo folders are visible in the selector.
The decision
Reader loading is reported separately. Warm reader inference is not the same as a cached-vector lookup.