I need a lightweight image-classification prototype that cleanly separates iPads from iPhones. You will work in Google’s Teachable Machine so the end result can be demonstrated live to non-technical stakeholders and, if needed, exported for further use in TensorFlow or a Python pipeline later on. Data I will supply a mixed set of photos—my own device shots plus carefully curated stock images—so the training set covers varied angles, lighting, and backgrounds. Target performance The prototype should reach better than 90 % accuracy on fresh, unseen images. Please incorporate any practical tricks (augmentation, class-balance tweaks, transfer learning, etc.) that help hit this benchmark without overcomplicating the workflow. Workflow & knowledge transfer Alongside the model, I need a concise walkthrough (screenshots or short Loom-style video are fine) showing: • how images were uploaded and labeled • which Teachable Machine options you chose and why • how to export the model and run a quick test Keep the explanation simple enough that a non-developer on my team can repeat the process or demo the model at a meeting. Acceptance criteria • A Teachable Machine project link or .zip export that achieves ≥90 % accuracy on my hold-out set • The walkthrough documentation described above • A short note on further accuracy-boost ideas should we want to push beyond the current scope If this sounds straightforward to you, let’s kick things off—I can share the first image batch as soon as we agree on timing.