Cover Photo

Inspiration

Medical education is still heavily based on passive studying. Students spend hours reading lecture slides, yet many struggle when they face their first patient encounter or OSCE examination.

We wanted to bridge that gap by transforming static lecture notes into an interactive clinical learning experience where students actively apply knowledge instead of simply memorizing it.

The vision behind MemoriaMD was simple:

Every lecture should become a patient you can learn from.

What it does

MemoriaMD converts a lecture PDF into a complete AI-powered learning session.

From a single uploaded lecture, the platform:

Generates a structured Teaching Blueprint
Creates a realistic virtual patient case
Simulates an interactive patient conversation
Evaluates the student's performance as an AI examiner
Produces personalized review questions and study material
Generates lecture-specific MCQs and flashcards

Rather than replacing educators, MemoriaMD helps students repeatedly practice clinical reasoning in a safe environment before interacting with real patients.

How we built it

The application follows a modular AI pipeline powered by the Gemini API.

Each stage of the learning workflow has a dedicated responsibility:

Teaching Blueprint generation from uploaded lectures
Clinical case generation
Interactive patient simulation
AI examiner assessment
Personalized reinforcement generation

To improve reliability, structured outputs are validated using Instructor and Pydantic, ensuring consistent JSON responses throughout the pipeline.

The application is built with:

FastAPI
React
TypeScript
Python
Docker

How we used the Gemini API

Gemini powers every intelligent component of MemoriaMD.

It is responsible for:

Understanding uploaded lecture material
Creating structured Teaching Blueprints
Generating clinically consistent patient scenarios
Simulating realistic patient conversations
Providing structured examiner feedback
Producing personalized review sessions
Generating lecture-specific MCQs and flashcards

Using Gemini's reasoning capabilities allowed us to create a coherent, end-to-end educational workflow rather than isolated AI features.

Challenges we ran into

One of the biggest challenges was maintaining consistency across multiple AI-generated stages.

Each component depends on the previous one, so preserving context while producing structured outputs was essential.

Designing reliable schemas, validating every response, and grounding each stage in the uploaded lecture helped make the workflow significantly more dependable.

Another challenge was balancing educational realism with development time, leading us to focus on a polished MVP that demonstrates the complete learning journey.

What we learned

This project reinforced that successful AI applications are built around thoughtful system design rather than a single prompt.

We learned the importance of structured outputs, modular pipelines, prompt engineering, and creating AI experiences that solve real educational problems.

Most importantly, we learned how powerful large language models become when they are grounded in a user's own knowledge instead of generating generic content.