AuditCheck.AI

Transparency First: Medical Bills, Decoded. Hospitals often hide costs until it's too late. We help you find the fair price and spot hidden charges before you sign the papers.

Comment 3

Inspiration

It started with a personal realization: medical bills are terrifyingly opaque. You get a piece of paper at the end of a hospital stay with vague line items like "Consumables" or "Procedure Charges" that seem to vary wildly from one hospital to another.

We realized that while patients trust their doctors with their health, they often don't have the tools to trust the financial side of healthcare. We wanted to build something that bridges that gap, a tool that doesn't just say "this is expensive" but explains why, using data to highlight fair market rates and AI to sniff out those sneaky hidden costs.

What it does

AuditCheck.AI is a transparency engine for healthcare costs.

Price Calculator: Users enter a procedure (e.g., "Total Knee Replacement") and their location. The app immediately compares the hospital's quote against government schemes (like PMJAY) and fair market private averages. It does not just look at the surgery fee, it warns users about specific "gotchas" associated with that procedures like unlisted "Implant Markups" or mandatory "Physiotherapy Packages" that bloat the bill.

Ask AuditAI: This is the core intelligence. Users can chat with an AI. Instead of generic advice, the AI receives the exact context of the user's selected hospital and procedure. You can ask "Is this stent upgrade charge normal for Mumbai?" and it answers based on our structured dataset.

How I built it

I built this as a robust full-stack application:

Backend: Node.js and Express formed the server and the backbone of AuditCheck.AI. I chose PostgreSQL for structuring and persisting user data.

The Brains (AI): I integrated Groq’s API to power the "AuditAI" chatbot. We needed speed, so we utilized the Llama-3.3-70b-versatile model. The breakthrough was passing a "Context Object" (JSON data containing specific hospital rates and hidden costs) silently in the background of every prompt. This prevents the AI from hallucinating and forces it to act as a strict auditor. Moreover I integrated Noupe AI chatbot named "Kiwi" to power AuditCheck.AI with other user questions related to the website or healthcare practices.

Frontend: I used EJS for server-side rendering to keep it paired with Tailwind CSS for a clean glass-morphism UI. I added GSAP animations to make the data visualization feel modern.

Open Source Architecture: I designed the codebase with a "Green Zone / Red Zone" philosophy. I separated the data layer "healthcare_pricing.json" so that open-source contributors can add hospital data for their cities without risking the stability of the core backend logic.

Challenges I ran into

The AI integration was the biggest hurdle.

Model Deprecation: Midway through development, the initial model (llama3-8b) threw 400 Bad Request errors. I had to quickly pivot to the newer llama-3.3 models and refactor our API calls to handle the change.

Context Injection: Getting the AI to give specific financial advice without being a financial advisor was tricky. I had to fine-tune the system prompt to make it act strictly as an "Auditor" that analyzes the provided JSON data rather than giving general medical advice.

Data Structure: Designing a JSON schema that could handle the complexity of different currencies, states, and hospital types (Government vs. Private) across different countries (India, USA, Argentina) required several iterations.

Accomplishments that I am proud of

Real-Time Contextual AI: The chat isn't just a wrapper for GPT. It knows the fair price for a knee replacement in the desired Country and State the moment you ask the question. That connection between static database data and dynamic AI reasoning is our proudest technical achievement.

Resilient Error Handling: I built a safe mode for the data loading part. If the pricing json fails to load or has syntax errors, the app degrades rather than crashing, ensuring the user always sees something.

What I learned

I finally learned how to integrate free ai models after several setbacks with other AI tools in the market and cracked the flaws I faced previously in rest of my work.

What's next for AuditCheck.AI

Community Verified Data: I plan on adding a "Trust Score" to the hospital data, allowing users to upvote or downvote cost estimates based on their actual bills.

Global Expansion: My data structure already supports multiple currencies. I want to actively recruit contributors from other regions to fill in the data map.

+ 1 more
Share this project:

Updates