Clinical.dev

Inspiration

The goal and inspiration was to try and speed up the clinical trial development process, especially with the use of emerging AI technology.

What it does

The app allows users to search for common endpoints/objectives/estimands for any disease using AI search and summarization capabilities. Additionally, a user can upload a draft/finished protocol pdf, get a PICOT summary of said protocol, and find other studies from clinical.dev that match the current upload to do a 1 to 1 comparison.

How we built it

The app was built using the NextJS framework on the frontend, with the backend being Flask utilizing OpenAI, FAISS, and Groq with Langchain for AI functionality.

Challenges we ran into

The biggest challenges were AI prompting and tuning for best results and confirming accuracy of said results. Additionally trying to cut down on the time required to produce AI summarized results. Another challenge is token limits. As of right now, the project does not support any protocols above a certain token limit (~approximately 10MB pdf), but is actively being worked upon.

Accomplishments that we're proud of

The link and comparison with clinicaltrials.gov was a good milestone and one we're proud of. Additionally, the swap to Groq from OpenAI has seemed promising.

What we learned

We learned how to utilize various AI technologies in a full end to end project

What's next for Clinical.dev

Increasing token limits for uploads, adding named projects with save functionality to reference in the future, and the generation of a draft protocol given user specifications.

Built With

• faiss
• flask
• groq
• langchain
• nextjs
• openai
• typescript