Billbuster

Home: Upload a photo or PDF of any medical bill, or try a sample. Files are never stored.
Results: The red pen marks every likely error. This ER bill has $295 in questionable charges.
Explanation: Every charge explained in plain English, so you know exactly what you're paying for.
Dispute letter: One tap writes a dispute letter citing the exact errors, ready to send
Pat call: Practice the call with Pat, an AI billing rep powered by ElevenLabs, before you call for real. One more thing: your homepage

Inspiration Medical bills are one of the most confusing documents people ever receive, full of cryptic codes, vague descriptions, and totals that are hard to verify. Billing errors like duplicate charges and inflated prices happen, but most people never check because they don't know how. We wanted to give everyone a tool that reads a bill the way an expert billing advocate would, for free, in seconds.

What it does

BillBuster turns a confusing medical bill into something you can understand and act on:

Snap or upload a photo or PDF of a bill or Explanation of Benefits
Extract every line item (code, description, quantity, charge) using AI
Flag likely errors: duplicate charges, quantity anomalies, prices far above typical ranges, and math mistakes
Explain each charge in plain English
Fight back with a one-tap dispute letter or financial-assistance request

The total questionable amount is calculated as:

$$\text{Questionable} = \sum_{i \in \text{flagged}} (\text{charge}_i - \text{expected}_i)$$

How we built it

Next.js + TypeScript + Tailwind for a fast, mobile-friendly web app
Google Gemini API reads bill images and PDFs and returns structured JSON, validated with Zod
A rule-based flagging engine in TypeScript. We deliberately kept error detection deterministic instead of relying on the LLM, so results are consistent and explainable
Gemini generates plain-English explanations and personalized dispute letters
[ElevenLabs voice agent that lets users practice negotiating with a billing rep]
[Tiger Data / Vultr, if used] ## Challenges we ran into
Bills vary wildly in layout, so getting reliable structured extraction took a lot of prompt and schema iteration
Deciding what the AI should do versus what code should do. We learned LLMs are great at reading and explaining, but flagging errors needed hard rules
Building realistic demo bills without using any real patient data
[Add a real bug or struggle your team hit; judges love honesty here] ## Accomplishments that we're proud of
A working end-to-end flow: upload → flags → explanation → dispute letter
Every flag comes with a clear reason, so users understand why something looks wrong
Privacy by design: uploads are processed in memory and never stored ## What we learned
How confusing healthcare billing really is, and how much money people may be losing to errors ## What's next for Billbuster Real price-transparency data from hospitals' published price files
Insurance plan awareness (deductibles, copays, in-network rules)
Tracking dispute outcomes so users can see what actually got refunded
Partnering with patient advocacy groups and community health clinics