Inspiration
We were inspired by the idea of running simulations using small llm agents to act out as individual entities. In industries like healthcare, where testing on real customers might have consequences and handling data requires a lot of care, simulations like these can save a lot of costs and uneccessry effort.
What it does
Our platfiorm takes in inputs of given scenarios through the form of files, form, documents to begin with. The user then defines the demographic distribution of the synthetic participants, as well as the actions they can take in relation to the triage documents. The user can then see and customize the system prompt an agent would be given, as well as quesitons for the agents to answer based on why they took each action.
After all the inputs are in, the platform simulates the agents interacting with the given documents, and is able to rank the triage based on the given metrics. The user would be able to interact with the results and see what demographic conditions influenced each agent to take any given decision. The emerging results at scale can then show how bigger trends emerge.
How we built it
We split the task by having Kyryl find and work with identifying real data for an NHS scenario, where he located form and demographic data in order to flesh out and confirm that running such a scenario would be possible.
Ozkan and Alex focused on developing the UI and frontend of the proof of concept demo.
Jacob did research into the concept and how to implement it
