MediMentor AI: Your Personalized Clinical Tutor

🧠 Inspiration

“What if every med student had their own AI mentor?”

We’ve seen firsthand how frustrating medical education can be — expensive, outdated, and sometimes disconnected from real practice. MediMentor AI was born from a desire to level the playing field and make expert-led learning accessible to everyone.

💡 What It Does

Personalized AI tutor trained on medical QA data

Adaptive certification assessments

Interactive 3D case simulations

Real-time mentorship matching with physicians

Think of it as your clinical co-pilot, designed to elevate both learning and confidence.

🛠️ How We Built It

Stack Layer Tools Used Frontend React, TailwindCSS, Three.js Backend FastAPI, Python AI Model Hugging Face Transformers, LoRA, QLoRA Inference NVIDIA Triton, HP AI Studio, vLLM RAG + Storage LangChain, ChromaDB, FAISS Experimentation MLflow, Weights & Biases (optional)

🚧 Challenges We Faced

Quota exhaustion on OpenAI — solved by training our own model

GPU limits on local hardware — overcame with HP AI Studio’s A100s

Fine-tuning large LLMs — used QLoRA for memory-efficient training

3D and AI sync — aligning visual case studies with AI responses took time

🌱 What We Learned

How to implement scalable RAG pipelines

Tuning LLMs using PEFT + LoRA with minimal VRAM

Building engaging UIs for medical learners without overwhelming them

Seamless multi-agent orchestration using LangChain

🚀 What’s Next

Deploy to mobile using React Native

Add voice interaction with Whisper

Onboard real mentors via verified profile linking

Clinical simulation in VR (stretch goal) 🎯

Built With

• chromadb
• falcon
• fastapi
• hp-ai-studio
• huggingface
• langchain
• llama
• medmcqa
• medqa
• mistral
• mlflow
• nvidia-triton
• pubmedqa
• python
• react
• transformers
• typescript