Copilot Studio Knowledge Assistant Agent
As this was an internal agent made for the company I've previously interned at, a lot of the finer details will be excluded for the sake of confidentiality. All information about this project will be high level and non-confidential.
For my internship at John Hancock, based in Boston MA, I was given a project to create an internal knowledge agent using Microsoft Copilot Studio. The purpose of this agent was to contain knowledge to help boost worker productivity and make gather information a less arduous process. The agent had connectors to our internal GitHub repository containing a classified business tool, access to the corresponding Confluence documentation, and access to an internal database for table lookups.
To begin with this project, I first had to get approval from other sectors, writing a mission statement that outlined the value that this agent would add to the business. There is no reason to create this agent if its just going to drain money without any upsides right? Once I created the mission statement and provided examples as to how the agent would behave, I was given approval to begin its development.
After the initial setup, detailing the model to be used (GPT 5.5 at the time of my departure) and basic security setup, connectors were set up to GitHub, Confluence, and our Azure SQL database for the agent to have the relevant information to operate.
Read access to our internal GitHub repository was given so that non-technical members on our team could ask a question about how codebase works and recieve a simple answer that they can understand with minimal programming knowledge.
Access to our Confluence documentation was also necessary as it detailed important information about the architecture of our project, which our agent could easily reference to give a better explanation.
Finally, we also gave the agent read access to our internal database so that it could lookup data values and spot or clear up errors and confusion. Connecting the database was a challenging task, as it required reaching out to employees from different teams to ensure that our use of the database would not prove to be a security risk.