ScaleDown Challenge Projects Divy – Health Insurance FAQ & Coverage Assistant
Introduction: Divy is an AI-powered multilingual assistant built to simplify the way people interact with health insurance policies. Insurance documents are often long, technical, and difficult to understand. Important details about coverage, exclusions, deductibles, and claims are buried inside complex legal language. Most policyholders only realize this when they need urgent medical assistance and cannot quickly find answers. Divy is designed to solve this gap. It converts complex insurance documents into structured, compressed, and conversational knowledge so users can simply ask questions and receive clear, reliable answers. The system supports policyholders, insurance agents, and policymakers through role-based guidance and multilingual communication.
Problem Background: In the current insurance ecosystem: Policy documents are lengthy and difficult to interpret Users depend heavily on call centres Coverage misunderstandings lead to claim rejections Agents spend significant time answering repetitive queries Language barriers limit accessibility These challenges create confusion, operational overload, and dissatisfaction among members.
System Architecture: Divy follows a structured workflow from document ingestion to intelligent response generation. Step 1: PDF Upload Users upload insurance policy documents in PDF format. Step 2: Document Processing The system extracts and cleans text content from the uploaded file. Step 3: Intelligent Compression Using ScaleDown, documents are compressed by approximately 85 percent while preserving semantic meaning. This reduces token usage and increases processing speed without losing context. Step 4: Knowledge Base Creation The compressed content is converted into embeddings and stored in a vector database. Step 5: User Interaction Users select: Their role Their preferred language They can then begin interacting with Divy conversationally. Step 6: Retrieval-Augmented Response The system retrieves the most relevant content from the knowledge base and generates context-aware responses using the language model.
Features: Insurance Knowledge Base – Divy extracts information from: Coverage terms Claims procedures Deductibles Waiting periods Pre-authorization rules Appeals process Instead of searching manually through documents, users can simply ask questions naturally. Document Compression – ScaleDown reduces document size by around 85 percent. Benefits: Faster response time Reduced computational cost Lower hallucination risk More accurate contextual retrieval This is one of the most critical technical components of the system. Coverage Checker – Users can verify: Whether a treatment is covered If waiting periods apply Deductible requirements Policy exclusions This reduces uncertainty during medical emergencies. Claims & Appeals Guidance – Divy provides structured guidance on: Claim filing steps Required documents Pre-authorization process Appeals workflow This ensures users follow correct procedures and avoid rejection due to incomplete documentation. Role-Based Access – Divy supports three types of users: Policyholder Insurance Agent Policymaker Each role influences how responses are generated. For example: Policyholders receive simplified explanations Agents receive operational details Policymakers receive policy-level insights This makes the system adaptable to different stakeholders. Multilingual Support – Divy supports: English Hindi Urdu Punjabi Telugu Bhojpuri Language selection happens before interaction begins, making the system inclusive and accessible. Conversation Memory – The system maintains structured conversation history: SystemMessage() HumanMessage() AIMessage() This ensures: Context continuity Consistent responses Better follow-up handling Reduced contradiction Conversation memory significantly improves reliability and user trust.
Expected Impact: Based on system design projections: 85 % document compression Approximately 70 % automated query resolution Around 55 % reduction in call enter volume Increased claims accuracy Improved member satisfaction Tech Stack: Python Streamlit ScaleDown Langchain Structured Memory Management
Future Scope: Future improvements may include: Real-time claim tracking integration Voice-based interaction WhatsApp and mobile deployment Insurance analytics dashboard Fraud detection module
Conclusion: Divy demonstrates how context engineering, document compression, and retrieval-augmented AI can transform traditional insurance support systems into intelligent, scalable, and user-friendly digital solutions.
Log in or sign up for Devpost to join the conversation.