Why Healthcare?

 

Perfect accuracies are not required in the recruiting industry. At the end of the day, when making a hiring decision, there will always be some human-in-the-loop looking at a candidate’s profile as a quality check to ensure fitness for a job. Frankly, in software-as-a-service (SaaS) generally, perfect accuracies are not required. Consider the dominance of large-language models (LLMs) in Internet search. Despite hallucinations and a non-deterministic nature, LLMs are being actively adopted and used (hopefully, in conjunction with efforts to mitigate their hallucinations). We personally like over Internet searches to be perfectly accurate but that doesn’t seem to have stopped large technology companies like Google pushing LLMs into their search products. It doesn’t really matter if there are inaccuracies in Google’s LLM-generated search summaries if there are also website links that we can visit to find the answer myself. (We use Google the way people used Google before LLMs—to find relevant websites.)

We like to aim delusionally high with respect to setting objectives while at the same time relying on science and data to make informed decisions on the way towards achieving those objectives. LLMs can be leveraged to clean and make sense of natural language data with perfect, reliable accuracy, as proven by our bulk resume parser product. Our larger goal with Vy Labs’ recruiting work is to develop a deep learning system that perfectly – and based solely on merit – can match candidates’ resumes to jobs. We believe by leveraging proper data cleaning/standardization enabled by LLMs, we can train and deploy better and more performant deep learning models. Our problem is that in the recruiting industry, perfect accuracy is not a necessity, even for a job-candidate recommender. Without the incentive to pursue perfection, how can we ensure that our organization is incentivized to develop systems that allow perfection?

Why Healthcare Matters for Perfection

The healthcare industry is our answer. Healthcare is an area where developing products is a long-term, slow burn of an investment. It’s also an industry in which taking research out of the lab into use in the real-world is tricky, particularly where data is necessitated like in deep learning. One reason for this is that AI-enabled systems in healthcare, such as clinical decision or diagnostic support, need to have perfect accuracy.

If we surface an FDA clinical trial or a medical protocol for a doctor that is relevant to a patient they are seeing, the doctor needs to trust that our system is accurate and its outputs verifiable. Even with a doctor-in-the-loop, inaccuracies in such health systems may cause misdiagnoses or otherwise adversely affect patient health.

Moreover, the healthcare industry faces serious constraints – especially in underserved areas – where delivering perfect performance is arguably more critical than anywhere else.

Underserved regions often struggle with:

1. Access: Under-resourced areas globally lack robust medical staffing and reliable internet connectivity. If you’re in a remote rural clinic, you can’t always depend on an internet-based solution.

2. Cost: Bleeding-edge technology is usually prohibitively expensive, and rural areas are rarely prioritized early in a tech product’s lifecycle.

3. Sensitive Domain: Healthcare, more than recruiting or typical SaaS, cannot tolerate hallucinations or missing links. A wrong recommendation might significantly impact patient health.

That’s precisely why we are pushing ourselves to produce perfect AI outputs: because healthcare demands it. And that internal drive for perfection will only enhance and refine our solutions in areas (like recruiting) where the tolerance for error is higher.

Dohrnii: Our AI-based clinical decision support tool

The same platform underlying our recruiting products also underlies our healthcare research work, specifically our efforts to develop an AI-assisted clinical decision support tool for primary care. We call it Dohrnii. Its purpose is to tackle the issues of access, cost, and sensitive diagnoses head-on:

1. Empower primary care in remote areas with an offline tool:

We’re leveraging LLMs and hybrid AI breakthroughs in ways that don’t rely on steady internet access.

2. Distribute Dohrnii at zero cost:

We aim for perfect performance on primary care tasks, but we also believe rural clinics shouldn’t bear any financial burden for cutting-edge AI.

3. Bridge resource gaps and uplift health standards:

Ultimately, Dohrnii will help primary care physicians (PCPs) quickly diagnose, triage, and determine next steps based on the latest protocols and research-backed guidelines—even in underprivileged areas.

Our Team’s Healthcare Background

Vish has a background in biomedical engineering, where he developed machine and deep learning models for symptom diagnosis. He spent his PhD refining data pipelines for messy patient data, collaborating with clinicians to ensure real-world impact. After earning his doctorate, he founded a digital health startup and led the ML team, building deep learning tools for automatic stroke symptom diagnosis for rural areas. He’s passionate about bridging scientific rigor and user-focused design, ensuring our AI systems truly empower clinicians and patients alike.

Our Chief Medical Officer, Dr. Dhanusha Subramani, is a general medical officer in the United States Navy. She attended Georgia Tech where she obtained a Bachelor of Science in Literature, Media, and Communication, and then attended The George Washington University School of Medicine and Health Sciences for her M.D. She currently serves as the primary care physician for a Marine Corps unit and deals with the complexities of the primary care ecosystem day in and day out.

A primary care physician is often both the first and last stop in a patient’s path: they’re crucial in the beginning of a patient’s medical journey and often remain the point of continued management for a definitive diagnosis. PCPs have scope over all healthcare specialties and are regarded as the “jack of all trades and master of none” since the breadth of knowledge required is enormous. When looking at medicine through this generalist lens, it can be difficult to leverage patient care against the vast expanse of a rapidly evolving knowledge base. AI-assisted clinical tools can help close this gap—and many more.

How Dohrnii Works

The first way that Vy Labs is addressing these problems is through Dohrnii, our clinical decision-making tool that helps triage patients, determine next steps in care, and assist in diagnosis and management based on current standards and research-backed guidelines. By using smaller parameter models, our larger goal is to allow broader implementation of AI-driven resources to underprivileged areas at minimal or zero cost. We’ve proven through our recruiting products that we can handle large-scale data with efficiency and keep costs low on our end—now we’re applying that same large-scale approach to an environment where 100% accuracy matters.

1. Parallelized LLM Calls & Info Retrieval

Our system efficiently makes parallelized calls to LLMs and includes a built-in document parser/answer engine. Recruiting taught us how to handle massive volumes of data; now we’re using that know-how to ensure medical protocols and clinical guidance are retrieved precisely, not guessed.

2. Supports Custom Deep Learning Models

We train and fine-tune smaller GPT-style models on biomedical data, aiming to eliminate hallucinations so that any doctor using Dohrnii can trust the outputs – even offline.

3. Edge Deployment for True Accessibility

Our short-term goal is to run Dohrnii entirely on edge devices without a persistent internet connection. That’s critical for underserved communities where connectivity is unreliable or nonexistent.

Looking Ahead

Ensuring our system can handle the complexity of healthcare tasks in a highly accurate way will naturally ensure that our platform performs efficiently in simpler SaaS tasks. Our overarching mission is to support physicians—particularly primary care providers—across different scopes of medicine, boosting healthcare productivity and ultimately enabling safer, higher-quality patient care.

We’re currently seeking partners to help us integrate and test Dohrnii in primary care clinics in rural South India. By tuning the system in real-world contexts, we aim to prove that perfect-performance AI tools can be both cost-effective and impactful, especially where they’re needed most.

That’s why healthcare.