Copilot

• 

  Slide

Inspiration

I recently booked an ENT appointment in the U.S. for the first time. It took me 2–3 hours to find a doctor who matched my preferences. Through research, I discovered that 23 million patients face similarly time-consuming searches (averaging 2 hours). 60% of them don’t schedule appointments due to the hassle. When they turn to call centers for help, the average wait time is 5 minutes, leading to a 7% abandonment rate and \$45,000 in daily losses.

What It Does

Copilot is an AI matchmaker that connects patients with the right doctors at the right time based on their preferences—saving hours and increasing hospital bookings.

How We Built It

We used machine learning models for embeddings and query restructuring, FAISS for similarity search, fuzzy matching for attribute scoring, and Flask to expose a REST API that recommends doctors based on patient queries.

Challenges We Ran Into

We initially attempted to build a scalable system using Docker Compose orchestration, but integration bugs slowed us down. We pivoted and deployed the system locally to meet the deadline.

Accomplishments We're Proud Of

We started as a team of three, but eight hours into the hackathon, one teammate left to pursue their own project. We quickly re-scoped and still delivered the project on time.

What We Learned

Ram and I learned a lot about each other. We discovered how to operate as a cross-functional squad to tackle product design, engineering, and delivery in tight timelines.

What’s Next for Copilot

We plan to pitch Copilot to the Commure team and explore how it could be integrated into their platform. Our hope is to make a meaningful impact on patients’ lives in hospitals across the U.S. and beyond.

Built With

• a
• and
• api
• faiss
• flask
• fuzzywuzzy
• json-for-data-storage
• numpy
• python
• rest
• sentence-transformers
• tensorflow
• transformers