🌍 Hack4IMPACT (Track 2) — Team P.cpp

Repository: Hack4IMPACTTrack2-P.cpp

Building high-impact, transparent Artificial Intelligence for real-world application.

---

👥 The Engineering Team: P.cpp

We are a collective of developers focused on building systems that matter.

Name	Role	Roll Number	Institutional Email
Prithanjan Acharyya 👑	Team Lead	$25051032$	25051032@kiit.ac.in
Piyush Ranjan Jha	Lead System Architect	$25051029$	25051029@kiit.ac.in
Prarabdh Mishra	ML Engineer (XAI)	$25051369$	25051369@kiit.ac.in
Pramesh Srivastava	Backend & Reporting	$25051368$	25051368@kiit.ac.in

---

🎯 The IMPACT: Why We Built This

For Track 2, we focused on solving a critical bottleneck in the deployment of deep learning models in high-stakes environments (like healthcare): The Black Box Problem.

High-accuracy models exist, but they lack interpretability. A user cannot trust a prediction model outputting $P(\text{Target}) = 0.98$ without knowing where the model is looking. Our solution bridges the gap between raw algorithmic power and human trust by providing Explainable AI (XAI) as a service.

---

🧠 System Architecture & Mathematics

1. Deep Learning Backbone

We employ a state-of-the-art Convolutional Neural Network (CNN) backbone. The model processes a tensor $X \in \mathbb{R}^{H \times W \times C}$ to produce a classification probability: $$\hat{y} = \sigma(\mathbf{W}^T \cdot \phi(X) + b)$$ Where $\phi(X)$ represents the high-dimensional feature vector extracted from the final global average pooling layer.

2. Explainable AI (Grad-CAM)

To solve the transparency issue, we implemented Gradient-weighted Class Activation Mapping (Grad-CAM). The importance weights $\alpha_k^c$ for a given class $c$ are calculated using the partial derivatives of the output w.r.t the feature maps: $$\alpha_k^c = \frac{1}{Z} \sum_i \sum_j \frac{\partial y^c}{\partial A_{ij}^k}$$ The final visualization heatmap is passed through a $ReLU$ activation to isolate positive clinical contributions: $$L_{Grad-CAM}^c = \text{ReLU}\left(\sum_k \alpha_k^c A^k\right)$$

---

🚀 Key Features

• ⚡ Sub-Second Inference: Optimized for rapid triage and high-pressure environments.
• 🗺️ Visual Localization: Dynamic heatmaps overlay over the original input to prove the AI's "thought process."
• 📊 Confidence Analytics: Multi-class probability distributions rendered via clean UI charts.
• 📄 Automated PDF Reporting: Instantly compiles the raw image, Grad-CAM heatmap, confidence scores, and user notes into a downloadable, professional report.

---

🛠️ Tech Stack Integration

• Frontend: React 19, Tailwind CSS 4, Recharts (for probability visualization), jsPDF (for report generation).
• Backend: Python 3.11, Flask, CORS.
• Machine Learning: PyTorch, Torchvision (ResNet50/DenseNet), pytorch-grad-cam, OpenCV (for tensor-to-image heatmap overlays).

---

⚙️ Quick Start (Run Locally)

1. Clone the Repository

git clone [https://github.com/Prithanjan/Hack4IMPACTTrack2-P.cpp.git](https://github.com/Prithanjan/Hack4IMPACTTrack2-P.cpp.git)
cd Hack4IMPACTTrack2-P.cpp