SafeScript

Analysis of conversation and patient's medical history
Conversation with SafeScript AI

Inspiration

Physicians today operate under extreme time pressure, and conversations are rarely analyzed after they happen. And once a diagnosis is made, there is often no structured mechanism to stress-test that decision. The system assumes the doctor is always correct. Today, patients are already seeking external verification. And currently, that verification happens outside the clinical system —unstructured, uncontrolled, and without medical oversight. We asked ourselves: What if that verification layer existed inside the healthcare system —designed responsibly, transparently, and as a support tool for physicians?

What it does

It works in five steps:

It records the doctor-patient conversation.
Converts it into a structured transcript.
Integrates uploaded medical history.
Performs AI-based clinical analysis.
Generates summaries, diagnostic critique, and risk flags. Finally, we provide an interactive chatbot where physicians can ask: “Why was this flagged?” This creates transparency in AI reasoning.

How we built it

Frontend: React Backend: Flask Cloud deployment: Vercel LLM: Gemma

Challenges we ran into

When creating the transcription feature on the website, we were originally going to run the entire OpenAI Whisper model to ensure the highest accuracy possible. However, this requires a significant amount of memory—likely much more than what the doctor's device has available. After noticing this problem, we researched other speech-to-text models online and found an open-source version of Whisper that maintains nearly the same accuracy while achieving a 127x speedup. After switching to this model, we found that the transcription process ran significantly faster, and cheaper.

Accomplishments that we're proud of

We built an AI-powered clinical safety layer designed to strengthen decision-making, reduce blind spots, and increase patient safety, and deployed it online.