Patronus AI

Bridging the Gap between Pharma Sales Reps & Doctors with AI!

Comment

Inspiration / Problem Statement

One of our team members’ mothers is a doctor who runs her own hospital. She spends nearly one hour, two to four times each week, summarizing and relaying the latest pharmaceutical information to her colleagues after hours. That recurring effort—and the risk of information loss it entails—sparked the idea for Patronus AI.

Physicians already dedicate over an hour per month—and up to several hours when you count follow‑up calls—to in‑depth pharmaceutical discussions, yet the insights from those conversations remain trapped in one‑on‑one silos. Manual note‑taking and fragmented sharing means every doctor re‑learns the same data, wasting time and risking patient care.

Additionally, clinicians engage in countless peer‑to‑peer discussions—case conferences, department rounds, informal hallway consults—where they share treatment insights and drug efficacy observations, but these rich dialogues often go undocumented or scattered across personal notes.

We propose Patronus AI: an AI‑first platform that records, transcribes, and semantically analyzes every pharma‑physician interaction and doctor‑to‑doctor discussion in real time, generating rich, searchable “Drug Summaries” and distributing them instantly across the network—freeing clinicians from manual documentation, democratizing critical drug insights, and giving them more time to focus on patients.

What it does

Patronus AI captures both in‑person and virtual meetings between pharmaceutical sales reps and physicians, then uses AI to:

  • Record and transcribe conversations in real time via OpenAI’s Whisper API (Basically a ground up Commure's Scribe)
  • Automatically extract key points, generate tags, and title each meeting with GPT‑4o
  • Link discussion topics to relevant clinical trials via the ClinicalTrials.gov API
  • Provide powerful, full‑text and tag‑based search across all summaries
  • Deliver daily audio summaries of key insights using OpenAI’s TTS API
  • Provides a custom Voice enabled AI agent 'Ask AI' to answer questions, or anything related to any summaries

Main Goals / Targets

  • Minimize manual effort for doctors and physicians as much as possible, such as writing detailed notes and sharing them with the appropriate stakeholders.
  • Simplify information consumption for doctors and physicians through human-voice audio summaries and a Q&A or doubt-solving AI agent.
  • Reduce information silos as much as possible while saving time for both doctors and pharmaceutical representatives.

How we built it

  • Frontend: Next.js, styled in TailwindCSS 4 and Heroicons for a responsive UI
  • Backend & Storage: Supabase (PostgreSQL) for data and audio-bucket storage
  • AI Integrations:
  • 1. Whisper for transcription
  • 2. GPT‑4o for semantic analysis (key points, tags, titles) and 'Ask AI' feature
  • 3. TTS‑1 for audio summaries
  • APIs: ClinicalTrials.gov for research links; Next.js API routes to glue everything together

Challenges we ran into

  • Setting Up Backend: Keeping up with sync of the flow, for example, recording audio -> saving it in db -> summarizing it correctly and getting relevant high-impact tags -> filling out the meeting summary card -> posting on dashboard. Had to work with lots of API calls and make sure they are triggered at the correct time.
  • Designing Scalable Solution: Engineering such a solution that should be easily scalable and reliable.
  • Ask AI: Making sure that the Ask AI agent has the latest context, when new meeting summaries are loaded, they had to be loaded into Ask AI's context.

Accomplishments that we're proud of

  • End-to-End Prototype: From recording, transcribing, summarizing, integrating advanced AI models, deploying on vercel, created a complete MVP.
  • Ready to integrate with Commure's Pre-existing Product: The transcribe page currently contains our custom made 'Scribe' like implementation. This can be easily exchanged with actual 'Scribe' to give more relevant and reliable outputs.

What we learned

  • Commure and Commure's Vast range of awesome products and services
  • Niche pain-points of integrating APIs and creating an end-to-end MVP and how to solve them
  • Prompt engineering is crucial and small tweaks to GPT‑4o prompts dramatically improved nearly all of the responses

What's next for Patronus AI

  • Login for doctors/medical staff using their Commure account
  • AI will use Commure's database of patients and tell how many current patients can be affected with the new drug
  • Mobile App
  • Analytics Dashboard

Built With

  • api
  • next.js
  • openai
  • supabase
  • vercel
  • whisper
Share this project:

Updates