#1 How We Started

27 September, 2023

Image: Out running at Waiuku Forest, 2022

Hi, this is Chester. Welcome 🫶. Thanks for your interest in this post, please read on. Briefly about me in a sentence, I am a former medical doctor turned software engineer and am the founder of Kekeno Tech - this small New Zealand-based business focused on software development in the area of healthcare and related research.

This is the first post in a series focused on people, activities and current topics that are pertinent to the work that Kekeno is doing. In particular, we will be exploring current state issues related to artificial intelligence and health IT in general in Aotearoa; as well as some fun technical, pure engineering topics. Please subscribe to our mailing list if you’re interested to stay updated.

To date, I haven’t been someone with a natural inclination for blogging, or putting myself out there on social media. We’re faced with an epidemic of click bait, reading headlines, subtitles and introduction paragraphs for our news; as our lives continue to move more into the digital space I suspect and fear this trend will only continue. For those of us who share this concern, what can we each do on a personal level to combat this? I have found spending more time reading longer articles, investigative journalism and books helps. And it is what is happening in the domain of my work and technical expertise that has inspired me to write¹.

My interests lie in health IT, and increasingly, AI applied to health. I have a network of friends, colleagues and collaborators with great skill and intelligence in the areas of software/computer science including AI, and clinical medicine. Yet it remains confronting how siloed we still remain across those domains of expertise, even now as AI (GenerativeAI, ChatGPT for example) storms ahead. In this post I won't get into the political and regulatory aspects of AI.

I recently presented on AI and machine learning to a group of students in year 3, MBChB University of Auckland. It was a great session with really neat engagement and discussion. About a week before that I caught up with two of my close friends (now specialists in Infectious Diseases and Intensive Care Medicine) from my own med school class. This week I had a meeting with some collaborators on current and upcoming research around a diagnostic feedback app for use in secondary and tertiary care. In all of these cases I was confronted with the same thought - there are many and complicated positive and negative aspects of AI applied to health (and of course more broadly beyond health) that we are not proactively working enough in Aotearoa and even the those of us who have hybrid expertise in this area are going to have to work hard to realise the benefits and to manage the risks. We shouldn’t be comfortable letting AI (in particular large language model-related work) proceed full speed ahead, driven by commercial interests and big tech. Clinicians/clinical researchers should be at the forefront of this work in terms of implementation, safety and risk management. 

It is heartening to see some signs of life here now. https://www.nejm.org/ai-in-medicine.

Those threads will be returned to in a subsequent post. I would like to finish introducing my personal journey to how I ended up in this vocational territory.

In 2015 I was working as a medical doctor (senior registrar in Respiratory Medicine) at Middlemore Hospital. I had no idea how computers worked, no expertise in programming and had no significant interest in artificial intelligence. My attention at the time was on clinical work, reading medical journals, areas like interstitial lung diseases, improving my procedure skills in bronchoscopy, thoracoscopy and EBU, reporting cardiopulmonary exercise tests, etc. I really enjoyed my work and 🫶 Middlemore Hospital.

In early 2016 I had some type of state change. I was becoming increasingly sensitised to clinician bias, observing frequent instances where this led to diagnostic or treatment errors and potential or definite patient harm. A few things to emphasise here - this is a general healthcare issue, not hospital specific, I have seen it across all hospitals I have worked at in NZ; we are not perfect, clinicians on the whole do a fantastic job and I have enormous admiration for my medical colleagues. However, despite this awareness, that was part of the inception of the idea in my mind - perhaps a computer program or AI would help protect against clinician bias. Little did I know that such a thought was not new and lots of work dating back even into that 80's had looked into that (see "Decision Support Systems")². 

Around that same time I had the increasing desire to understand computers and learn to program. I watched a few movies where the protagonist (usually a benevolent hacker) was at a linux terminal powering the computer to his/her own will, doing something that seemed almost magical. I started to wonder how the future would look in 10 (it's almost been 10 at the time of writing....💀), 15+ years from now. I felt that I would be at a disadvantage and relatively disempowered if I didn't understand computers in this coming future context.

I continued my clinical work, had a few more run-of-the-mill disillusioning experiences (clinicians among us all know the various shapes and sizes these come in) and so after one such experience in particular I made a bold decision to stop full time clinical work and go back to university to study computer science - specifically, I quit. This was pretty bold and I think most of my colleagues thought "he'll be back". I was 29 at the time. A couple of weeks after making this decision I sat in a university mathematics tutorial doing differential equations - something I hadn't done for a good decade or more!

Over the ensuing 2.5 years, I completed my BSc in computer science ✔ ️. For the first ~ 2 years of the degree I worked part time as a locum medical doctor doing ~ 30 hours per week with the full time university program. Year 3 of the degree was more challenging and I wouldn't have been able to (in a non-healthy way) do well in the courses as well as satisfactorily complete my medical CME. So (sadly!) I decided not to renew my annual practising certificate.

I then went on to complete the Honours year, taking papers in AI, machine learning, advanced algorithmics among others. I began to focus on conversational AI, including for my dissertation. On a whim one night I created a chatbot using generative AI (recurrent neural network model training on ~ 5 million size Reddit data set) - it was pretty good though limited compared to current LLMs (transformer based models). I worked part time with the HABITs (now called Tirohia) research team as a software engineer and as a Research Fellow at the University of Auckland. I had come a fair way since 2016; quite a bit further in fact than I had intended to go but was loving working in computer science. I felt cognitively and creatively stimulated and empowered. I had achieved what I wanted to achieve but had sacrificed active clinical work - which I do miss (like an ex partner who you still love and care for but have moved on from!).

The academic environment is fantastic for pure research but not optimal for agile software development. I experienced some limitations in the pace and culture of building software in that environment, and together with some exciting private project opportunities I began to take on private work. This was something which happened organically.

Kekeno was incorporated in 2021. I was (and am) lucky to be in contact with other talented software engineers and designers. I saw that there was this void where clinicians and researchers had great ideas for projects but no providers/collaborators of the right size, agility, expertise to cover the software development side. And so we got started. We’ve so far been involved in some really neat, inspired projects and continue to branch into more and exciting new projects and research.

Stay tuned for the next post, we’ll be diving into interesting and pertinent topics in this space.

Please send any feedback to chester@kekeno.techChester

‍Footnotes, References and Resources
‍1. Written without the use of GenerativeAI 
2. These systems and the early (early AI general approach) using symbolic representation, a knowledge base and logic rules are interesting in their own right and we will for interest give attention to this in the future.