FIX-R (AM-AI)
AI-Powered Visual Diagnostic Platform
Developed a full-stack AI platform that diagnoses issues in homes, vehicles, and marine systems using computer vision and hierarchical machine learning.
-
Built a hierarchical ML pipeline that routes images through domain → primary → secondary classifiers to determine accurate diagnoses
-
Implemented a hybrid inference system combining on-device CoreML models (privacy-first) with cloud-based TensorFlow/Keras models for enhanced accuracy
-
Designed a confidence-based decision engine to resolve model conflicts and assign severity/urgency scores
-
Developed an image preprocessing pipeline (HEIC/RAW → normalized tensors) with EXIF correction and adaptive contrast enhancement
-
Integrated real-time object detection (COCO-SSD via TensorFlow.js) to validate predictions and dynamically adjust confidence scores
-
Built support for AR-assisted repair workflows and step-by-step guidance
-
Deployed backend infrastructure to AWS (EC2, RDS, S3, SageMaker) to enable scalable cloud-based processing alongside on-device inference
-
Implemented a multi-stage ML pipeline capable of dynamically selecting models based on the input domain
-
Designed system to process and analyze user-submitted images in near real-time
-
Integrated multiple AI validation layers, improving prediction confidence through cross-model verification
-
Built a hybrid system supporting both offline (on-device) and online (cloud-enhanced) operation