AidAura

User details page
Cost breakdown
Financial options
Home Page

Inspiration

For the past seven years, one of our teammate's grandparents have been in and out of hospitals. And every time, their family wasn't just worried about their health, but rather they were worried about the bills. Even with insurance, they never fully knew what was covered, what assistance programs existed, or whether they were making the smartest financial decision possible. They'd spend hours on hold with insurance companies, googling terms they didn't understand, and hoping they weren't missing something important.

Millions of Americans face the same impossible maze every time they walk into a hospital. The data that could help them exists, but nobody puts it in front of them when it actually matters. That frustration is what inspired AidAura. Because guessing costs too much.

What it does

AidAura takes the guesswork out of medical bills. Users simply upload their insurance card or enter their details manually, and AidAura does the heavy lifting — automatically pulling their coverage information and showing them personalized ways to save based on their actual plan and hospital.

If you're insured, AidAura finds every benefit you're entitled to right now. If you're not, it figures out which charity care programs and financial assistance funds you qualify for and connects you directly. No hold music. No confusing paperwork. No hoping you didn't miss something. Just clear, actionable answers at the moment you need them most.

How we built it

AidAura is built on a Next.js frontend and Flask backend, with OCR handling automatic extraction of key coverage details from uploaded insurance cards. To map a patient's described situation to the right billing codes, we built a vector database using Action Vector DB to identify relevant CPT codes, which then feeds into a pricing engine that calculates in-network and out-of-network cost estimates in real time. The AI decision-making layer runs on the Groq API, processing the patient's insurance, location, income, and household size to surface personalized financial pathways — whether that means finding every applicable benefit for an insured patient or identifying qualifying charity care programs for an uninsured one.

Challenges we ran into

Data acquisition and structuring — There is no clean, ready-to-use dataset, so we spent significant time parsing complex CSV outputs to extract and structure the medical code information we needed.
Symptom-to-CPT code mapping — Translating natural language symptom descriptions into the correct CPT codes proved harder than expected, since the same condition can map to dozens of codes depending on context and severity.
Insurance card parsing variability — Determine how differently insurance plans structure their coverage information made it difficult to build a reliable extraction layer that worked consistently across card formats and plan types.

Accomplishments that we're proud of

Building something that actually matters — It is something that we have experienced ourselves and so have millions of Americans, we're proud to have used this experience to create a product that delivers real financial guidance at the moment people need it most.
Turning unstructured data into intelligence — We successfully sourced, generated, and structured usable healthcare data from scratch, then built a pipeline on top of it that connects CPT codes, insurance coverage, and charity care eligibility into a seamless real-time experience.

What we learned

Healthcare pricing is highly fragmented and inconsistent. Working with CPT codes exposed how costs can vary based on network status and negotiated rates, requiring us to design logic that accounts for that variability.
Strong AI depends on well-structured data and retrieval. We learned how to normalize clinical data, implement hybrid semantic search, and tune filtering logic so cost predictions were reliable and context-aware.
Simplicity is critical in crisis design. In emergency scenarios, minimizing user input and reducing cognitive load isn’t just good UX — it directly impacts clarity, trust, and decision-making.

What's next for AidAura

Direct insurance integration — Real-time API connections with major insurers so coverage data is always current and automatic.
Bill negotiation assistance — Helping patients identify errors and overcharges on itemized bills and guiding them through the dispute process.
Broader accessibility — Expanding our charity care database and adding multilingual support so every patient can access the help they qualify for.

Actian

Our backend leverages Actian, a vector AI database, to power medical code retrieval in two stages. First, we store 39 high-level medical code categories in the vector database, and when a user describes their visit, Groq selects the most relevant categories as an intelligent filtering layer. We then perform a deeper hybrid semantic search across 1,164 detailed procedure codes — using vector embeddings and cosine similarity — filtered by those categories to identify the most likely procedures the patient will undergo.