Onco Mind

Leverages Google Cloud AI, Gemini AI, and BigQuery ML to deliver cutting-edge cancer treatment recommendations. Our platform continually learns from clinical outcomes to improve accuracy and efficacy.

Comment

Inspiration

OncoMind was born from a deeply personal place. One of our team members lost a close one to cancer, and the journey through complex treatment options, inaccessible clinical trials, and scattered information was overwhelming. We realized how many others face this uncertainty—and how AI can help bridge the gap between patients and the right care. OncoMind is our way of turning grief into purpose and technology into empowerment.

What it does

OncoMind is an AI-powered clinical trial matching and patient analytics platform designed for oncologists, researchers, and healthcare providers. It:

  • Uses LLMs to match patients to eligible clinical trials
  • Allows natural language descriptions of patients to find best-fit trials
  • Extracts structured features from unstructured eligibility text using Gemini AI
  • Provides secure JWT-authenticated login for admins and physicians
  • Offers an intuitive dashboard to manage patient profiles and view recommendations

How we built it

Our stack combines the best of AI, cloud, and modern web:

  • Frontend: React.js with Vite for fast SPA performance
  • Backend: Flask with REST APIs, JWT Auth, OAuth (Google), and mail support
  • AI Models: Gemini Pro for structured data extraction
  • Database: Google Cloud SQL (PostgreSQL) for secure patient and trial storage
  • Data Ingestion: BigQuery pipelines that preprocess trial eligibility and patient data
  • Deployment: GitLab CI/CD for automated GCP deployment (Cloud Run, GCS hosting)

Challenges we ran into

  • Handling unstructured eligibility criteria was harder than expected—most trial data was noisy and inconsistent.
  • OAuth flows and redirect URIs created deployment edge cases between local and production.
  • CORS issues when deploying the frontend to GCS and the backend to Cloud Run.
  • Keeping inference latency low despite using external APIs and vector databases.

Accomplishments that we're proud of

  • Successfully built a full-stack clinical trial recommender system powered by real patient data and AI
  • Engineered a robust BigQuery + Gemini AI pipeline to structure messy trial data.
  • Created a secure and beautiful dashboard for real-world clinical users
  • Made our project personal, meaningful, and impactful.

What we learned

  • Best practices for full-stack CI/CD deployment using GitLab and GCP.
  • Translating clinical jargon into structured features requires not just AI—but empathy.
  • Importance of modularizing services and blueprints in Flask for scale.
  • That building something meaningful beats building something flashy.

What's next for Onco Mind

  • Integrating Patient Chat Agent: A conversational AI to help patients understand trials
  • Adding Multilingual Support: Supporting trial matching in Arabic, Hindi, and more.
  • Fine-tuning the eligibility model with more clinical datasets from TCIA and NCI.
  • Onboarding real oncologists and hospitals through outreach and early pilots.
  • Submitting to journals or conferences to publish our method of AI-based eligibility extraction.
Share this project:

Updates