rishikumar37282-creadevity/Public-Griverance-Project-IIT-Mandi

AI triage for public grievances — a BiLSTM + Attention network that reads citizen complaints and predicts department + urgency (100% Critical-recall), with a citizen portal, PyTorch · FastAPI · zero JS dependencies. # ⚖️ Jan Samadhan ### Public Grievance Urgency Classification System **Track:** AIML · **Domain:** Public Service & Civic Operations · **Difficulty:** Intermediate · **Duration:** 10 Days A complete, production-ready AI system that reads citizen complaints in plain English and instantly predicts the responsible department, urgency level (Low/Medium/High/Critical) with an explainable 0–100 urgency score, and generates a concrete escalation and action plan — ensuring serious issues reach officers faster instead of waiting in a first-come-first-served queue. --- ## 🚀 Live Deployment | Application | Live URL | |---|---| | 🏛️ **Citizen Portal (Jan Samadhan)** | public-griverance-project-i… | | 📊 **Model Observatory (Admin)** | public-griverance-project-i… | | 🔬 **Live Model Explainer** | public-griverance-project-i… | | 📖 **Project Story Deck** | public-griverance-project-i… | > ⚡ Deployed on **Render** (free tier). The service spins down after 15 minutes of inactivity — the first request may take **30–60 seconds** to wake up. --- ## 📋 Table of Contents - Phase 1: Problem Understanding & Stakeholders - Phase 2: Data Preparation & Weak Supervision - Phase 3: Model Architecture — BiLSTM + Attention - 3.1 Layer-by-Layer Architecture Breakdown - 3.2 Embedding Layer - 3.3 Bidirectional LSTM - 3.4 Additive (Bahdanau) Attention Mechanism - 3.5 Context Vector & Weighted Sum - 3.6 Multi-Layer MLP Heads - 3.7 Multi-Task Learning: Two Parallel Heads - Phase 4: Data Preprocessing Pipeline - 4.1 Tokenization - 4.2 Vocabulary Building - 4.3 Sequence Encoding & Padding - 4.4 Train/Val/Test Split (Stratified) - Phase 5: Training Configuration & Hyperparameters - 5.1 Complete Hyperparameter Reference - 5.2 Loss Function Design - 5.3 Class Weighting & Critical Boost - 5.4 Optimizer & Scheduler - 5.5 Early Stopping & Checkpoint Selec …