a link for the article was added
The BG-AppealAI project is an example of how vector search and generative AI can be used in applications developed using InterSystems technologies. The developed application is designed to assist in writing appeals against an insurance company’s refusal to pay for hospital procedures.
You can use git or zpm to install our project.
git clone https://github.com/banksiaglobal/AppealAI.git
docker-compose up -d
zpm "install bg-appealai"
In both cases, after installing the project, create a new .env
file in the root folder with the following variables:
OPENAI_API_KEY="YOUR_API_KEY"
PATH_SAVE_DOC="/opt/irisapp/"
and create credential as described in Step 2: Set up OpenAI Operation of the article Tutorial: Adding OpenAI Interoperability to Production.
The frontend application is available at the link:
http://localhost:32783/appealai/index.html
You can test our app in two ways: through a demo of our app or with any web client like Postman.
In the root of the project you can find the file AppealAI.postman_collection.json
, which you can import into Postman and test the application API. Please note that the queries in the collection are arranged according to the logic of performing actions inside the system, so for correct testing they should be executed sequentially.
The main functionality of the backend part of the application is implemented using Interoperability Production. After successful installation the project locally, you can go to the Production with the following url: Production
_SYSTEM
SYS
The Production consist of the following elements:
The functioning of the system can be seen most clearly in the visual traces.
Visual trace of taking an insurance contract, dividing it into vectors, and storing the vectors in a database (is invoked by requesting “upload insurance contract” from the postman collection, execution time ~ 30 minutes):
Visual trace of receiving a rejection letter, sending requests to OpenAI, and saving the appeal letter if an appeal is possible (is invoked by requesting “upload rejection letter” from the postman collection, execution time ~ 1-2 minutes):
All business processes in the system are implemented as BPL, which helps to understand the logic “behind the scene”:
We are the team of Banksia Global company and we always are opened for something new and interesting!
#BanksiaGlobal #LangChain #VectorSearch #GenerativeAI #AI #Angular