FairClaim AI – Unbiased Multimodal Claim Adjuster

Google Solution Challenge Prototype Theme: Unbiased AI Decision – Ensuring Fairness and Detecting Bias in Automated Decisions

FairClaim AI (BiasLens Auditor) is a state-of-the-art, multimodal AI application designed to detect, flag, and mitigate human bias in automated insurance claim models. By natively processing both visual evidence (vehicle damage) and audio statements simultaneously, it audits "Legacy" biased models, flags demographic unfairness (accent, gender, background), and generates an objective, corrected claim decision.

Inspired by industry-leading end-to-end multimodal architectures (like Gemma 4 E2B mockups), this prototype is powered by Google Gemini 2.5 via FastAPI.

🚀 Features (The WOW Factor)

1. Multimodal Native Processing: Upload an image of damage AND record an audio statement. Gemini 2.5 processes both simultaneously.
2. Real-time Bias Auditing: The system actively simulates a flawed "Legacy Model", runs a "Bias Detection Engine" to flag unfairness, and then outputs a "BiasLens Corrected Decision", proving exactly how the bias was fixed.
3. Built-in Audio Recording: Directly record the driver's statement in the browser using the MediaRecorder API.
4. Stunning Auditor Dashboard: A visually striking, glassmorphism-inspired interface featuring a 3-card comparative audit report perfectly tailored for a high-impact hackathon presentation.

---

🛠️ Tech Stack

• Backend: Python FastAPI, python-multipart, Google Generative AI SDK
• Frontend: HTML, JavaScript, Tailwind CSS (via CDN), FontAwesome
• AI Model: gemini-2.5-flash

---

📦 Step-by-step Setup Instructions

1. Prerequisites

• Python 3.9+ installed
• Google Gemini API Key (Get one at Google AI Studio)

2. Backend Setup

1. Navigate to the backend folder:

   cd backend

2. Create a virtual environment:

   python -m venv venv
   venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Set up your environment variables:
   • Create a .env file inside the backend folder.
   • Paste your Gemini API key: GEMINI_API_KEY=your_api_key_here
5. Run the backend server:

   uvicorn app:app --reload

   The backend will start at http://127.0.0.1:8000.

3. Frontend Setup

1. Simply double-click the frontend/index.html file to open it in your web browser. (Note: To use the microphone recording feature, modern browsers require the site to be served over localhost or https. If the microphone button doesn't work via file://, simply start a quick local server in the frontend folder using python -m http.server 3000 and visit http://localhost:3000)

---

🎥 The Winning Demo Script

Context: "Human claim adjusters can unconsciously bias their decisions based on a driver's accent, gender, or perceived socioeconomic status over the phone. FairClaim AI eliminates this."

Step 1: Open the stunning Dark Mode UI. Step 2: Drag and drop an image of a damaged car (e.g., a rear-end collision). Step 3: Click the Record button under the Audio section. Speak into the microphone with a specific accent or tone: "Yeah, I was sitting at the red light and this guy just slammed right into the back of my car." Click Stop. Step 4: Click "Generate Fair Claim Report". Step 5: The UI will load and display the dual dashboard.

• Show the Statement Consistency box: Point out how it transcribed and verified the story against the physical image damage perfectly.
• Show the Fairness Check box: Highlight the text where the AI explicitly states it ignored the demographic/accent markers to ensure an unbiased decision.
• Toggle to the JSON tab: Show the judges the raw, perfectly structured JSON output, proving it's enterprise-ready.

Conclusion: "FairClaim AI doesn't just automate claims; it enforces fairness at the architectural level using Google Gemini's multimodal capabilities."