Vishnu Rachakonda is a Machine Learning engineer at Tesseract Health, a 4Catalyzer company focused on retinal imaging. In this role, he builds machine learning models for clinical workflow augmentation and diagnostics in on-device and cloud use cases.
Outside of work, he is the operations lead for the ML-Ops Community, a 3500-person online hub of ML-Ops practitioners and enthusiasts, and co-hosts the ML-Ops Coffee Sessions podcast.
Summary
In this episode, Vishnu starts off narrating his day to day as an ML engineer in a Healthcare startup. We then delve into what’s going well for AI in Healthcare. The realist in Vishnu explains what’s not going well with AI in Healthcare with emphasis on whether AI is helping to bridge the clinical gap in knowledge.
The focus then shifts to novel geographies and how solutions to local healthcare problems would be best sourced locally from homegrown companies rather than come from big pharma or big tech. Given Vishnu’s passionate engagement with ML-Ops, we discuss how this field and community is maturing.
Audience Questions
[00:36:30]
Improving technology is the wrong goal. Sure it helps but the benefit is so small compared to the institutional healthcare problems that drive the slow speed and high cost that patients experience.
What is your take on this?
Ed Steck — Senior Technology Consultant, USA
[00:38:00] I would like to know how a technologist can associate themselves with dynamic trends and developments in the fields of ML which are growing at an exponential pace, especially when they have to pick the technology stacks for their product?
Sagar Deyagond — Senior Design Engineer @ BD, India
[00:41:50] For companies utilizing/developing AI-based diagnostics: I wonder how patients generally respond to knowing the fact that an AI-based algorithm has categorized them as ‘high-risk’ vs. a human?
Rahul Sharan — Biotechnology Student at VIT, India
Topical Questions
[00:07:13]: As a machine learning engineer and you’ve looked at various problems from different angles, overall what do you think is working well with respect to the use of AI in healthcare?
[00:08:53]: What are the problems that you see and what do you think sort of ails the system of AI in healthcare?
[00:16:11]: Is it actually translating to clinical outcomes or is it actually translating to making impacts on patient’s lives, that remains to be seen. So, what would be your understanding of that?
[00:23:32]: Do you believe that there is perhaps a greater opportunity and an easier route to adoption of AI based solutions in these novel geographies, in developing countries rather than the developed western world?
[00:26:35]: If there are big and interesting problems that are worth solving, which will really address sort of the global total addressable market of healthcare, they will be done by big pharma or big tech or a combination of both.
So, that’s one argument I’ve heard and I’d be interested to know what you think of it?
Quotable Quotes
[00:6:00]: If Dev-Ops is where software engineering meets IT, ML-Ops is where software engineering meets machine learning.
[00:10:31]: When a clinical gap in knowledge exists, where does AI fit into that? Can we learn from AI? Is there a framework where we can learn from AI or are we basically just going and say until this clinical gap in knowledge is resolved, we cannot permit AI to function in this paradigm.
[00:13:49]: …There are a lot of things that we do in medicine that we don’t know exactly why they work…I think sometimes you make decisions based on what the results that you can see, even if you can’t fully explain them.
[00:20:11]: …We would never put in a chemical engineering plant, a machine that we didn’t understand because there’s risk there. But in ML, we are essentially putting out machines that aren’t as well characterized and that too in healthcare where risk is really not acceptable.
[00:29.21]: If you’re a local company and you understand how your society works and you understand the specific thing that makes a product work better than in your context than others, you can use that knowledge and build a different model of care.
[00:29.59]: I’m way more bullish, long-term on some very interesting homegrown companies developing and solving the problems in India than I aim that any big tech or big pharma would seriously make an indent in those problems.
[00:34:01]: 90% of the problems in ML engineering and data science come from people writing bad code or from following bad coding practices.
Notable Mentions
[00:05:15] ML-Ops Coffee Sessions Podcast
[00:33:57] Chip Huyen
[00:40:54] Clayton Christensen – The theory of jobs to be done
[00:46:31] ML-Ops Community Slack Channel
Pointers to past You+AI Podcast episodes
[00:14:48] S2-E5 Navigating AI in Medical Research, with Chris Lovejoy
[00:20:37] S2-E9 Putting the brakes on infectious diseases with AI, with Reshma Suresh
Connect
LinkedIn: Vishnu Rachakonda
Website: healthcareai.dev
The You+AI Vodcast
A companion video segment full of fun and candid moments.
Check it out here!