Skip to main content

What happens when two developers with a passion for healthcare and explainability team up for a high-stakes hackathon? You get OpenMed—a promising agentic AI system that helps make diagnostic insights more transparent and accessible for clinicians.

We caught up with Yeswanth Bharadwaij and Sriharsha “Sunny” Patallapalli, the minds behind OpenMed, to hear more about their journey through the HP + NVIDIA Hackathon and how they built their winning prototype using HP AI Studio.

 

💡 The Spark Behind OpenMed

Sunny, a former data engineer at GE Healthcare and a current Master’s student, had seen firsthand how difficult it can be for doctors to trust black-box AI systems. When he teamed up with Yeswanth, a data analyst with a finance background, the two found common ground in solving a critical pain point: how can AI support medical decisions in a way that clinicians understand and trust?

That question led to OpenMed, an interpretable AI assistant designed to work with medical imaging data. By focusing on transparency, explainability, and workflow integration, the duo aimed to build a foundation that clinicians could use without second-guessing the results.

 

“Doctors need to trust the system—not just get a prediction,but understand how it was made.” – Sunny

 

🛠️️ How It Works

OpenMed integrates multiple tools to build a complete, explainable diagnostic pipeline:

  • Custom Convolutional Neural Network (CNN) models: Based on two research papers and fine-tuned in HP AI Studio.

  • FastAPI: To expose the model as a service.

  • OpenWebUI: A sleek, ChatGPT-style interface that doctors can use to interact with the AI agent.

  • MLflow: Used for model tracking and deployment—new territory for Sunny, and a big value-add during development.

  • HP AI Studio: Served as the training ground for the model and a central hub for experimentation and collaboration.

“We hadn’t used MLflow before—HP AI Studio helped us integrate it easily and made our model management smooth.” – Yeswanth

 

🚀 Built in 5 Days

After some early brainstorming and idea refinement, the bulk of the OpenMed system was built in the final five days of the hackathon. Sunny focused on the model development and setup inside HP AI Studio (running on an NVIDIA GPU-enabled ZBook), while Yeswanth owned the frontend and agent routing logic. Integration came together just in time for submission.

Even with the time crunch, they prioritized explainability in the user experience, adding clear visual cues and model outputs that could support medical decision-making.

 

💻 HP AI Studio Highlights

For Sunny and Yeswanth, working in HP AI Studio was a key part of building and managing OpenMed efficiently. One of the standout features for the team was how seamless it was to collaborate—once the project environment was configured, it could easily be shared and pulled down by other team members, eliminating a lot of typical setup headaches.

Another game-changer was the built-in MLflow integration. Sunny, who had never used MLflow before, found it intuitive and powerful. He was able to register models, track experiments, and move toward deployment-ready workflows faster than expected.

 

“I’ve always used TensorBoard before, but using MLflow through HP AI Studio was incredibly smooth. I’ll probably keep using it going forward,” Sunny shared.

 

These built-in tools made experimenting, iterating, and syncing progress across the team much easier—especially during a fast-paced sprint like a hackathon.

 

🧭 What's Next for OpenMed?

The project is currently hosted locally, but Yeswanth and Sunny plan to:

  • Expand support for more diseases and imaging types.

  • Host and deploy the system for broader testing.

  • Seek funding and explore healthcare compliance paths.

  • Gain access to diverse and representative datasets to mitigate model bias.

As Sunny put it, OpenMed isn’t just a demo—it’s the start of something that could genuinely help bridge the gap between AI systems and clinical trust.

 

🔗 Dive Deeper into OpenMed

 

Be the first to reply!

Reply


HP AI Creation Center | HP Z

Terms & ConditionsCookie settings