This personal essay draws from a discussion with 34-year-old Pratik Desai from New Jersey, who developed a tool through a process called vibe coding. He later enhanced this tool with advanced coding techniques to assist his mother in coping with Stage 4 duodenal adenocarcinoma, an advanced form of cancer affecting the small intestine. The following narrative has been edited for brevity and clarity.
I have no formal medical training, but I have always been closely connected to the healthcare field due to my family’s involvement in it. After completing my studies, I landed a position at Accenture, where I specialized in systems integration and business analysis, exploring how technology impacts workflows, particularly in government-based medical and non-medical benefits. This led me to a strategy consulting role, primarily working for Johnson & Johnson. Additionally, I created an app to help my wife keep her medical credentials current.
Eventually, I joined Salesforce, marking a significant foray into AI. I progressed to become the global practice lead for personalization, which involved machine learning. Afterward, I founded an AI company called 1to1, which was acquired by ListEngage, where I now hold a leadership role.
I also initiated a podcast to educate people about AI and its potential applications in various aspects of life, including caregiving and health. Despite the surge in information that AI offers, a gap still exists in understanding how to effectively utilize it.
My mom was a completely healthy woman
On October 20 last year, she hosted a 70-person Diwali celebration. Little did we know that a month later, she would be diagnosed with Stage 4 duodenal adenocarcinoma, after experiencing stomach issues and urging her to visit the hospital.
Initially, the plan was to discharge her without scheduling an oncology appointment, which felt dismissive and indicated a lack of urgency. We were left with a diagnosis and a message of “good luck.”
Being a type A personality, I refused to accept that as the final answer. Having never been a caregiver before, I realized I had to find tools that could help us through this fight.
We decided we were going to fight — and AI would help us
My first step was to consult AI to become well-informed about Stage 4 duodenal adenocarcinoma and understand the challenges ahead. AI started coaching us and clarifying the doctors’ jargon.
Once we decided to battle the illness, I employed AI for extensive research on every oncology center in New Jersey. Despite Thanksgiving fast approaching, I contacted numerous facilities to secure an appointment within the week.
To my surprise, AI helped us discover a doctor on a hospital’s oncology webpage who turned out to be my high school ex-girlfriend. She quickly arranged an appointment for us.
My mother spent 76 days in the hospital, 67 of which were in patient care. Regrettably, at no point did we feel that the medical system prioritized her desires as a patient, particularly the need to have enough time to say goodbye to loved ones. Instead, it often felt like the focus was merely on progressing to the next step in her treatment plan.
At one instance, critical decisions were made based on a CAT scan. Our process uncovered two misdiagnoses in the scan results alongside three instances where the report wrongly identified the type of cancer. Reading that report ourselves would have been impossible, but AI enabled us to analyze it effectively.
The workflow I developed consisted of a daily export from the Epic system, inputting that data into NotebookLM along with any symptoms I noted or that my mother mentioned. I would instruct NotebookLM to “synthesize the data.”
Afterwards, I turned to my preferred language model, Claude, and asked for interpretations. Questions such as, “What should I prepare for in tomorrow’s appointment?” or “What second opinions should I request?” became commonplace.
The workflow wasn’t perfect to start
What we created is still a work in progress—far from perfect but functional and free.
My aim was to keep it straightforward, but as my mother’s medical records grew to 1,600 pages—excluding images and scans—this created challenges. Eventually, the volume of data became overwhelming, and I reached the limits of context processing.
Initially, I used Google AI Studio effectively, but eventually transitioned to Claude as its models improved. I aimed to utilize the best technology, especially with limited time remaining with my mother.
Throughout the 76 days of her illness, I remained by her side from 5 a.m. to 10 p.m. daily, and NotebookLM became an invaluable second opinion and source of guidance.
Some health professionals expressed skepticism about my approach, arguing that AI can yield inaccuracies and is only correct about 70% of the time. I countered by asking why the medical field isn’t held to the same standard. We expect perfection from AI, yet the healthcare system often falls short.
I believe the AI saved her life three times
At least three occasions arose where this workflow appeared to be lifesaving.
On Christmas day, I noticed subtle changes in her walking, breathing, and speech, along with her hesitance to interact with family. The medical team was notably unresponsive that day.
I input my observations into AI, which indicated she could be facing a pulmonary embolism. While I didn’t rush her to the hospital immediately, I did reach out to my cousin, a doctor, who advised me to seek emergency care. Had I waited for a call back from the medical team, it might have taken several hours.
AI also detected a troubling pattern of bleeding following two blood transfusions. We learned that transitioning her from a liquid to solid diet 48 hours after the transfusion irritated her ulcer, leading to significant bleeding. In these instances, we acted quickly, likely preventing critical consequences.
AI moved the needle for my mother
By asking the right questions and advocating for her care, we could ensure she received the time needed to say her goodbyes. She could share precious moments with my two-year-old daughter.
Many of my peers are dealing with similar situations in hospital settings, and I’ve shared this workflow with several friends. One of them felt that his mother’s healthcare team was lacking; he adopted this system, educated himself, and successfully advocated for her care.
In one notable instance, he convened a meeting with her doctors, impressing them with his understanding of the case without referencing any notes. They frequently asked if he had a medical background, to which he responded, “No, I work in marketing.” The power of this workflow truly makes a difference.
