MedELI5 | Devpost

The MedELI5 logo features a minimalist, clean design with soothing colors, symbolizing clarity and healthcare.

Inspiration

Our team has an extensive background of working on data science and data engineering problems in the healthcare domain. This gave us unique insights into some of the pain-points that exist in the current healthcare systems. One such problem is of how difficult it is for a patient to understand their various lab reports or answer their queries when they don't have immediate access to a doctor. We were also inspired by the internet trend of 'Explain Like I'm 5' which helped us come up with a name for the project!

What it does

The generative AI agent utilizes the extensive medical knowledge of LLM models such as Gemini 1.0 to explain various medical reports shared with the agent by a patient. Currently, our agent focuses on blood reports. The agent can explain the same report at different levels of medical expertise as well.

How we built it

We used Vertex AI Agent Builder UI to build our agents for this project.

Challenges we ran into

Maintaining consistency in AI responses and varying the response length was very challenging.

Accomplishments that we're proud of

We were able to complete a working Proof-of-Concept that showcases how useful such an agent can be. We were also able to achieve fairly good modularity and extensibility in our prompts.

What we learned

How we can utilize few-shot training to improve our LLMs performance

What's next for MedELI5

By leveraging the modular and extensible nature of our prompts, we can build multiple specialized agents that can explain various medical reports for various knowledge levels.