AI-Driven Insurance Claim Processing System

Insurance claims are still checked largely by hand: an adjuster reads scanned forms, compares them with the policy, looks for signs of fraud and then decides, which makes settlement slow and uneven. This paper presents the AI-Driven Insurance Claim Processing System, a role-based web application that automates the routine part of this workflow and keeps the adjuster in control of the rest. Policyholders submit a claim and upload supporting documents; optical character recognition (OCR) extracts the key fields; rule-based checks confirm that the policy is active and covers the claim; and a supervised machine-learning model estimates the probability of fraud and reports the factors behind it. A transparent weighted decision score built from document quality (30%), policy validity (30%), fraud safety (25%) and amount reasonableness (15%) routes each claim to an approval recommendation, adjuster review or investigation, and hard rules override the score. Every decision is stored with its component scores and reasons for audit. The system is built with React, FastAPI, scikit-learn and PostgreSQL. We describe the architecture, database design, REST interface and scoring algorithm, demonstrate the routing with a worked example, and discuss limitations and the planned evaluation.