Syed Tabish Shah's Work | ContraWork by Syed Tabish ShahSyed Tabish Shah
AI Engineer building intelligent chatbots, RAG systems, and
The Intel Scene Classifier is a end-to-end computer vision web application designed to classify natural scenes across six distinct categories: Buildings, Forest, Glacier, Mountain, Sea, and Street. Built around a fine-tuned MobileNetV3Small deep learning architecture, the application achieves a 90.47% test accuracy while maintaining lightweight inference times suitable for cloud deployment.
To enhance transparency and reliability, the system integrates Grad-CAM visual explanations, heatmapping the exact regions of an image the neural network focuses on during inference. Additionally, it features experimental Out-of-Distribution (OOD) detection to identify and handle unsupported image uploads cleanly. An experimental case-management and evidence-analysis application using YOLOv8 object detection. A FastAPI backend and React interface support evidence uploads, annotated results, timelines, reports, and chain-of-custody workflows.
Its GitHub Actions CI/CD pipeline tests and builds the React frontend, builds and scans the Docker image with Trivy, then uses passwordless Azure OIDC to publish to Azure Container Registry and deploy to Azure Container Apps. Built an Azure-native RAG medical assistant that answers health questions using WHO documentation on anaemia, diabetes, and hypertension. Pipeline: WHO PDFs → OCR → Gemini embeddings → Azure AI Search for retrieval → Gemini generates grounded answers with inline source citations. Includes a clean chat interface and a medical-disclaimer banner for responsible use.