Smooth Operator: A Healthcare Copilot

We provide a comprehensive suite of services to streamline hospital back-office operations, encompassing appointment scheduling, real-time transcription, document compilation, and claims processing.

Comment

Inspiration

In 2023, the US government spent $24 billion specifically on administrative expenses. Doctors can use smooth operator to automate call scheduling, live notes scribing and soap notes generating. Most importantly, smooth operator can be used to generate billing codes and automatically fill insurance claims on the doctor’s behalf, which is a very repetitive and time intensive process.

What it does

Automatically generating billing codes and filling insurance claim pdf forms by using patient details, appointment transcript, soap notes, xrays and supporting document summary.

This copilot also does automated appointment scheduling, live audio transcription and diarization, soap note generation, billing code generation and image summarization.

How we built it

We use automated appointment scheduling using Claude AI. We also use OpenDental as our backend database system.

We do live audio transcription and diarization using DeepGram.

Soap notes are generated google gemini. We use few-shot prompting to generate soap notes for the appointment.

We use GPT to generate billing codes and image summarization.

We use google gemini to interpret and generate the final fields which need to be filled in the insurance form claim. We use few shot learning and prompt engineering to generate the final data.

Challenges we ran into

Our biggest challenge was making sure that the LLM Model is not filling incorrect data into the claim form.

Also making sure that the soap notes are correctly generated from the audio and there is no inaccurate information. We also tested multiple models to see if they are accurately able to summarize dental xray images.

Accomplishments that we're proud of

We are able to fill dynamic insurance forms upto 80% with accurate information.

We are able to generate billing codes correctly which are crucial for filing insurance claims.

We are able to schedule appointments for patients based on availability with zero human intervention.

We talked to actual dentists to verify image summarizations generated by chatGPT and soap notes are able to summarize the context accurately.

We are able to generate real time transcription with accurate speaker diarization.

What we learned

How to use Google Gemini, Claude, ChatGPT and DeepGram API. We also learnt more about Flask and React and NextJS integrations.

What's next for Smooth Operator

Improving prompting to generate better information and giving model more context when filling insurance form.

Built With

Share this project:

Updates