Design Document ‐ Backend

Case Management AI - Backend Design Document

1. Introduction

This design document outlines the proposed design for the backend of the Case Management AI program.

2. Overview

The Case Management AI project is a cloud-based web application designed to assist caseworkers in managing client cases and providing personalized intervention ideas and support services. The backend application will provide case managers and clients with data storage, machine learning backed recommendations and the ability to design a custom recommendations based on the most accurate, up to date, data.

3. Technology & Architecture

• Back End (API): Will handle the communication between the front end and the machine learning model. Will use the Node framework.
• Machine Learning Model: This will be the core logic that processes data and returns results. Models will be created using RandomForest. RandomForest was chosen due to its ability to account for possible overfitting of the sample data, the databases used for the actual model will depend on our ability to gather data and track client outcomes.

4. Project Guidelines

This project will be build modularly, with every feature that the user can access from the front end being connected to an API that will access the machine learning model that calculates the results. Each feature will have an accessible endpoint. We will work with Beam Group to determine what a useful result looks like, whether it is a certain result or a certain intervention. We will also track outcomes for clients.

5. User Stories

*The social worker has access to the information the client input *The client has access to their own form *The client receives confirmation of receipt of their form to their email address *The case manager can make edits to the form if needed *The information is confidential between client and case manager *The case manager and client will both have access to the client's current return to work percentage *The case manager will be able to test out different combinations of interventions and override program recommendations

6. Deliverables

*API Specification Document *Endpoint Code *Authentication & Authorization *Model Code *Model Data *Unit Tests *Integration tests

7. Testing

*We will run tests with certain cases in which we should expect certain outcomes, and compare those to the results that are actually being produced by the model. *C