Have you ever noticed a spot on your skin and wondered if it could be something serious? Wondered if your cough indicated COVID or just the run of the mill flu? Wish you could check your blood pressure with your smartphone? Enter Helfie.ai, an emerging healthcare startup in the healthcare sector who are harnessing multli-modal AI algorithms designed to help patients quickly and accurately check for a range of healthcare conditions. With teams in Australia and Denmark, Helfie.ai aims to make healthcare more accessible and affordable by enabling patients to analyze their health conditions from the comfort of their homes, using only their smartphone.
Helfie.ai is building a solution that caters to healthcare providers who can arm their patients with an application to more quickly and accurately access the care they need Helfie.ai’s AI model analyzes pictures, videos, sound recordings through a patients’ mobile device to check for indicators of potential health conditions. Some of the diagnoses they support are respiratory health, vitals, and skin cancer. The product supports real-time doctor assistance to confirm the diagnosis and prescribe the appropriate treatment.
All assessments provided by Helfie’s AI are tentative indications of a possible condition, not an actual diagnosis. These assessments are under no circumstances a substitute for personal consultation, screening, diagnosis or treatment by a registered health practitioner. Helfie apps should not be used to make decisions in situations that are life-threatening.
Helfie.ai is part of the Microsoft for Startups Pegasus program, which provides access to Azure credits, Go-to-Market (GTM), technical support, and unique benefits such as access to Azure AI infrastructure on a dedicated GPU cluster.
Helfie.ai is leveraging a dedicated NDm A100 v4-series cluster in Azure ML to train and finetune the AI model that power their solution. Because of the need for clinical grade predictions about a patient’s health condition, their flagship product, Vital’s AI, analyzes images, video and sound recordings to evaluate health conditions and make remote assessments.
By using the dedicated GPU cluster of A100s, Helfie.ai’s CTO Nikhil Sehgal, reports that by using the dedicated GPU cluster of A100s, they’ve reduced their model training time from two weeks to just two hours.
“The drastic reduction in model training time is critical for a startup like us, it allows us to build faster and get our product into the hands of users,” Sehgal tells Microsoft for Startups.
The GPU cluster also allows them to scale up their models as they grow their user base and data sources, ensuring that their products can handle the increasing demand and complexity. Running workloads on Azure also helps them seamlessly comply with compliance policies like the Health Insurance Portability and Accountability Act (HIPAA) and General Data Protection Regulation (GDPR).
Helfie.ai has developed specialized AI models, known as “AI Specialists,” fine-tuned for specific medical domains. For instance, the “Cardiology AI” is an LLM (Large Language Model) with comprehensive knowledge in cardiology, enabling users to receive expert-level responses similar to those from a professional cardiologist. Helfie.ai accomplishes this by collecting reputable literature and documentation from the web to fine-tune both open-source LLMs and flagship GPT-series models.
To create and monitor these models, Helfie.ai utilizes Azure ML Studio for machine learning experiments, hyperparameter optimization, and performance reporting. The company relies on Azure Kubernetes Service (AKS) for intensive computational tasks, Azure OpenAI for multimodal LLM capabilities, and Azure AutoML for optimizing hyperparameters. They ran both classification and computer vision model experiments on Azure ML. Through these experiments, Helfie.ai was able to optimize hyperparameters including ideal n_estimators and max_depth.
Helfie’s assessments often involve visual or audio inputs, demanding remarkable computational bandwidth to train their workloads. A significant challenge was re-training the skin cancer prediction model using smartphone images of lesions. To address this, Helfie.ai established its entire training pipeline on Azure’s dedicated GPU cluster, improving throughput per epoch as new data from academic institutions necessitated frequent re-training.
To address data privacy, Helfie implements robust AI content safety policies to ensure user interactions with specialist LLMs are safe and remain confidential. This includes Azure AI Content Safety Dashboard, a real-time dashboard that flags responses with potential risks, helping to prevent any harmful user actions.
Helfie.ai’s products—Vitals AI, Respiratory AI, STI AI, and Skin AI—are now available on Azure Marketplace.
Dedicated GPU clusters are available for startups on Microsoft for Startups who are backed by one of Microsoft’s strategic VC partners: Microsoft expands free Azure AI infrastructure access to startups | Microsoft for Startups Blog, along with Startups who are active in the Pegasus Program.
