This essay is based on a conversation with Pratik Desai, 34, based in New Jersey. vibe coded tools — then refined using advanced coding tools — to help mothers Navigating Stage 4 Duodenal adenocarcinoma, an advanced cancer that affects the small intestine. His words have been edited for length and clarity.
I have no medical background. My family is in the medical field, so I have spent my entire life in close contact with the medical field.
After graduation, I took a job at Accenture, focusing on systems integration, business analysis, and the intersection of how technology impacts workflows, primarily working on medical and non-medical benefits at the government level. I moved into a boutique strategy consulting position with Johnson & Johnson as a major client. I also created an app to help my wife remember to keep her medical certifications up to date.
I found my way to Salesforce. This was my first real foray into AI. I have worked hard to become a global practice leader in personalization, which is synonymous with machine learning. I started my own AI company called 1to1, which became ListEngage, where I currently work as part of the management team.
We’ve also launched a podcast to introduce people to AI and what it can do for care, health, and all other aspects of life. AI is providing people with more information than ever before, but there are still gaps in people’s understanding of how to use it.
my mother was a perfectly healthy woman
On October 20 last year, she hosted a Diwali party for 70 people. Little did I know that a month later she would be diagnosed with stage 4 duodenal adenocarcinoma. We urged her to go to the hospital as she was having stomach problems.
The doctors’ plan was to discharge her without an oncology appointment, but we felt that this left her with little time and it was clear that the doctors were going to fire her. We basically received this diagnosis and the message, “Good luck.”
I’m a Type A person, so I wasn’t going to take that as an answer. I had never been a caregiver before. I don’t know what would happen without AI. I grabbed the tools in front of me and asked, “What can I do now?”
We decided to fight – and AI will help us
I started by asking the AI to make me an expert on stage 4 duodenal adenocarcinoma to find out what we were facing. The AI started coaching us and synthesizing what the doctor was saying.
Once we decided to fight, we used AI to thoroughly investigate every hospital in New Jersey. Even though it was Thanksgiving, I called every number I could and tried to schedule something for the week.
It was through the AI that I found a doctor on the hospital’s oncology department webpage who happened to be my ex-girlfriend from high school. She got back to us and set us up with an appointment.
Mother survived for 76 days, 67 of which were in the hospital. There was never an instance where I felt like the medical community actually cared about her goals as a patient, primarily about setting an optimal lifespan to say the goodbye she wanted to say. Instead, it felt like they were always looking at how quickly they could take us to the next step.
At one point, we were making important decisions based on CAT scans. The workflow identified two misdiagnoses on the CAT scan report and three instances where the report listed the wrong cancer. Neither my mother nor I would have been able to read the report, but the AI could read it and investigate it more deeply.
The workflow I ended up creating is one that actually works over time and is exported daily from the Epic system. We entered it into NotebookLM along with the symptoms I saw and what she told me. For NotebookLM, I would say, “Please synthesize the data.”
I then went to my preferred LLM (Claude) and had him explain what the information meant. Ask questions like, “What should I know about tomorrow’s promise or the scenario in front of me?” Later, I went back and asked. “What should I ask for a second opinion? What doesn’t seem right here that I need to push back on?”
Starting the workflow wasn’t perfect
What we’ve built isn’t perfect or elegant yet, but it’s accessible and free.
We wanted to keep things as simple as possible, but as her medical records reached 1,600 pages without images or scans, they could no longer be loaded into a single thread. In the end, there was too much data and no context.
Initially, we used Google AI Studio, which worked fine, but as Claude’s model improved, we eventually stopped using it. I knew I only had a little time left to spend with my mom and wanted to use the best model possible.
For the 76 days she was alive, I was at her bedside from 5am to 10pm every day, and NotebookLM was my second opinion and coach.
Some doctors I talked to about my product countered that AI can cause hallucinations, and that this is only true 70% of the time. I said, what if we graded our health care systems the same way? We expect AI to be perfect, but that is not the case in the medical world.
I think the AI saved her life three times.
I’ve had three emergencies based on this workflow and I believe it saved her life.
On Christmas Day, I noticed something was wrong with the way she walked, breathed, and spoke. She was refraining from spending time with her family. The health system did not respond to calls on Christmas.
I input what I was seeing into the AI and it determined that she was definitely dealing with complications from a pulmonary embolism. I didn’t rush to the hospital right away, but thanks to that I was able to send an SOS to my cousin who is a doctor and he told me to take him to the hospital. Emergency calls could take four to five hours to get a call back.
The AI was also able to detect a pattern in which she started bleeding seven days after the transfusion. We realized that 48 hours after the IV we would be taking her off the liquid-only diet and moving her onto food. This would irritate the ulcer she had and encourage it to bleed. The amount of blood she lost would be so severe that it would be virtually fatal. I caught this twice.
AI moved the needle for my mother
We were able to ask the right questions and encourage doctors to make improvements. By doing so, we were able to give her the time we needed to say goodbye. She was able to kiss her 2 year old daughter.
I’m at an age where many of my friends’ families are in hospital systems, so I’ve enabled this kind of workflow for some of them. One of my friends used this workflow to learn and get to the top of her mother’s health care team because she felt they weren’t doing the right thing.
In one instance, he called a meeting with doctors, who were completely impressed that he understood the entire case without a single note in front of them. They kept asking if he was a doctor or a biology major, but he said he was a marketer. That’s the power of workflow.