Project Title: Causal Multimodal Diagnostic Agent (Medical AI) – Code + Frontend + Research Paper Budget: 8,000 (Fixed Price) Deadline: December 20, 2025 (Strict) Project Overview: I am looking for an experienced AI/ML developer and researcher to build a Causal Multimodal Diagnostic Agent (CMDA). This system must integrate medical imaging (Chest X-rays) and clinical text reports to diagnose diseases, using causal graph learning to eliminate spurious correlations. The project requires delivering a fully working codebase, a basic interactive frontend for testing, and a complete, high-quality research paper suitable for publication. Key Technical Requirements (Based on Project Design) Multimodal Inputs: Image Encoder: ResNet50 or Vision Transformer (ViT) for Chest X-rays. Text Encoder: ClinicalBERT or BioBERT for radiology reports. Causal Fusion: Implementation of a Graph Neural Network (GNN) to model causal dependencies between image and text features. Explainability: The system must output: Grad-CAM heatmaps for X-ray visual explanation. SHAP/LIME scores for text importance. Dataset: Train on MIMIC-CXR (primary) and test robustness on PadChest or NIH ChestX-ray14. Deliverables A. Working Code & Frontend Complete Python source code (PyTorch/TensorFlow). Basic Frontend (Gradio/Streamlit): A web interface to upload an X-ray image and paste a clinical report to see: Predicted Disease Label. Confidence Score. Visual Heatmap (Grad-CAM) & Text Highlights. Instructions/Readme to run the project locally. B. Research Paper Full paper drafting (Abstract to Conclusion). Target: IEEE JBHI or Medical Image Analysis standard. Plagiarism: Must be under 10% (strictly checked via Turnitin). Must include comparative analysis tables and the Causal Attribution Graph. Milestone Breakdown (Total: 8,000) Milestone 1 (1,600): Data & Encoders – Preprocessing MIMIC-CXR and implementing Image/Text encoders. Milestone 2 (2,400): Causal Model & Training – Implementing the GNN Causal Graph, Fusion Layer, and getting initial validation accuracy. Milestone 3 (1,600): Frontend & Visualization – Creating the Gradio/Streamlit UI with Grad-CAM and SHAP integration. Milestone 4 (2,400): Final Paper & Packaging – Complete research paper (<10% plagiarism), final code cleanup, and project handover.