Inspiration
Dental insurance has one simple problem. You don't find out what something costs until you're already in the chair. Lincoln Financial already has a mobile app and a web portal with pieces of this, like a cost estimator and claims tracking, but they only answer you when you go looking. Nobody tells you ahead of time that you have $400 in benefits about to expire. We wanted to build something that looks out for you instead of just waiting around for a question.
What it does
Lincoln DentalConnect is an AI agent that sits on top of a dental plan. The two core features are still being finalized, but right now it centers on two things. Ask the AI lets you type a question in plain English, something like “where can I go within five miles that charges less than $300 for a teeth whitening or a filling”, and it works out a real answer using your plan and nearby providers. That's just a placeholder example for now, the agent is meant to handle any question in that style, not only that one. Find a Dentist shows a map of nearby dentists marked as in network or out of network, so you have that information up front before you choose someone, instead of finding out after the fact. It also isn't built just for Lincoln's own staff. Lincoln provides its dental network to a lot of other companies too, so we built this so any employer on Lincoln's network could use it under their own branding, not just Lincoln itself.
A few more pieces we've added to the dashboard, as separate features rather than one bundle:
• The annual maximum shown on the dashboard is $2,000. • Dependents show up automatically on the dashboard when we already have their information, with a plus button to add one by hand if needed. • Out of network care requires pre-authorization. Picking a specific out of network clinic triggers a pop up that gets the price pre approved and sends it to Lincoln Financial for final approval. • In network care gets a similar pre estimate, so an agent looks up the expected cost for that specific clinic and service ahead of time. • A payment plan agent asks someone what they're comfortable paying and over how many months, sends that proposal to the doctor for approval, and shows it on the dashboard once approved. • Login works as single sign on across Lincoln Financial and its client companies, so employees from any company on the network can sign in through the same system. • An agent can request or add emergency PTO for a doctor's appointment right from the dashboard, without leaving the app. • Suggested products, similar to how a shopping site makes suggestions, surface relevant product links after a procedure or diagnosis. This is not a food delivery feature, just product suggestions
How we built it
We built the screens, including the dashboard, Ask the AI, Find a Dentist, and Care Plan, so the look and color can change per employer instead of being hardcoded to one company. The AI layer runs on a model routing platform similar to OpenRouter, so we can call multiple AI models instead of being locked into just one, using function calling so the agent can decide when to look up a dentist, check benefits, estimate a cost, or book an appointment. We also brought in AWS Bedrock agents to handle the price estimation and pre-authorization workflows, working alongside the rest of the conversation. Deployment is on Vercel.
Challenges we ran into
Figuring out login for multiple companies was harder than it looked. Since Lincoln works with many employers, we couldn't just build login around Lincoln alone, so we had to plan for a setup where any employer's own login could eventually be used. We also had to be honest about what the AI doesn't know for sure. An agent that confidently gives someone the wrong coverage number is a real problem, not just a small bug, so we made sure the cost estimate clearly says it's an estimate, shows where the numbers came from, and reminds people that Lincoln makes the final call.
Accomplishments that we're proud of
We're proud that we turned separate steps, finding a dentist, checking if they're in network, and seeing the real cost, into one flow that actually ends with booking an appointment. We're also proud that we built trust into the design from the start instead of adding it later, since the estimate shows its sources and clearly says it needs confirmation. On top of that, we actually built the app so it works for more than one employer instead of just saying it would in theory.
What we learned
We learned a lot about how function calling works across different models, including some tricky details around keeping track of state across a conversation. We also got hands on with AWS Bedrock agents for the price estimation and pre-authorization pieces. And we came away understanding that planning for more than one company has to happen early, rather than getting bolted on at the end.
What's next for Public Static Void Main
Next, we want to connect to real claims and benefits data instead of test data, add real login support so any employer's workers can sign in with their own company accounts, connect booking to real dentist calendars, and look into what it would actually take to get this inside Lincoln Financial's own employee app.
Built With
- amazon-web-services
- boto3
- claude
- cloudform
- docker
- dyanmo
- fastapi
- framer
- mapbox
- node.js
- pydantic
- pytest
- python
- reactome
- typescript
- uvicorn
- vercel
- vite
Log in or sign up for Devpost to join the conversation.