About

We have a Flutter app that already integrates MediaPipe hand landmarks, but the configuration may be incorrect. We need an experienced mobile ML engineer to audit & fix the current pipeline and then add a fast, on-device nail segmentation model (stored locally, fine-tunable offline) that runs smoothly on iOS. Audit & Stabilize MediaPipe Verify graph/config parameters (model versions, confidence thresholds, smoothing, handedness). Fix bugs in platform channel / plugin setup if any (iOS focus). Add metrics (FPS, CPU/GPU %, dropped frames, landmark confidence). Output: short audit report + PR with fixes. 2) Nail Segmentation (On-Device) Recommend a lightweight segmentation approach (e.g., MobileNet-U-Net / DeepLab-lite or equivalent) suitable for Core ML or TensorFlow Lite with GPU/ANE acceleration. Implement ROI cropping using hand landmarks to minimize compute and improve mask quality. Export & integrate the model locally (no external calls). Provide a repeatable fine-tuning pipeline (offline): scripts + instructions to retrain with new labeled data and re-export to Core ML / TFLite. Expose clean Flutter APIs returning nail masks (and optional overlay preview). 3) iOS Performance Tuning Use Core ML / Accelerate / Metal (or TFLite Metal delegate) for inference. Optimize pre/post-processing, batching, quantization (INT8/FP16), and input sizes. Add runtime fallbacks (device capability checks) and graceful degradation.