InsureAI – AI-Powered Insurance Claim Evaluator

Overview

InsureAI is an intelligent insurance claim evaluation system designed to bring clarity and automation to one of the most frustrating areas of insurance: policy comprehension and claim assessment. By combining semantic search, vector databases, and large language models, the system can detect the type of claim, retrieve the most relevant clauses from lengthy documents, and evaluate eligibility with reasoning that balances both flexibility and explainability.

The inspiration for InsureAI came from a personal experience. I noticed how overwhelmingly complex insurance documents are, especially for older generations who often struggle with dense and technical language. Even reading a policy word-for-word does not guarantee clarity, as exclusions, conditions, and waiting periods are buried in jargon that can leave people feeling helpless. When I tried to help a family member navigate this process, I realized that there had to be a way to use AI not just to automate, but to empower. InsureAI was born out of that frustration, with the goal of making insurance simpler, more transparent, and accessible to anyone regardless of technical background.

Features

At its core, InsureAI automatically identifies whether a claim relates to medical, accident, or travel coverage. It then searches through the uploaded policy documents using FAISS and high-quality embeddings, retrieving the most relevant sections with precision. By combining these results with a hybrid reasoning engine—part LLM and part rule-based logic—the system delivers eligibility evaluations that are not only accurate but also explainable. The output is structured in JSON, containing the decision, the reasoning behind it, and references to the clauses that informed the outcome. To further improve accuracy, documents are divided into logical sections such as definitions, exclusions, and benefits, ensuring context is preserved during retrieval.

Tech Stack

The system is built on FastAPI for handling backend requests, FAISS for efficient semantic search, and SentenceTransformers for embeddings. Gemini API serves as the primary large language model, with OpenAI as a fallback to ensure reliability. For document parsing, PyPDF2 and python-docx allow the system to handle common formats like PDF and Word, making it flexible enough for real-world use.

Learnings

One of the most important lessons in building InsureAI was that simple approaches to document splitting did not work. Moving to section-aware chunking greatly improved retrieval accuracy and highlighted how critical preprocessing is in AI pipelines. I also discovered that a hybrid system combining LLM reasoning with deterministic rule-based checks provided the best balance between adaptability and trustworthiness—something particularly valuable in high-stakes fields like insurance. Beyond technical lessons, the project reinforced the importance of designing with empathy. Many of the challenges in insurance are not just computational but human, requiring systems that explain decisions clearly to non-technical users.

Future Scope

While InsureAI already automates claim evaluation and simplifies complex documents, there is much room for expansion. One direction is to build conversational assistants that can explain insurance clauses in plain language, adapting explanations to the user’s level of understanding. Another is to enable side-by-side policy comparison, helping users choose the plan that best suits their needs without wading through endless documents. For older users, voice-based interaction could make the system more natural and accessible.

Looking even further, I envision extending the system into a more immersive digital experience. By integrating a Three.js-powered visualization layer, InsureAI could present policy decisions in interactive 3D spaces. This extension would also serve as a platform to showcase my 3D Blender artwork and games, blending functionality with creativity. In doing so, the project moves beyond traditional insurance tech into a space where clarity, interactivity, and design converge to redefine how people experience digital tools.