The problem PocDoc solves

Healthcare is inaccessible. Patients can be far away from healthcare facilities, experience long wait times to access the services they need, and might not be able to afford healthcare. But with PocDoc, patients are now able to have access to quality healthcare because PocDoc provides mobile clinics and dispatch services for medical access to come directly to you. PocDoc reduces triage time since it can recommend to healthcare providers possible diagnosis and what medical services a patient needs before reaching the patient. PocDoc reduces patient wait time because it filters out the services patient might require and where to find them. This is a win-win solution for both parties. PocDoc also can greatly increase rural health care by providing the logistics to deploy mobile clinics/ what kind of clinics to certain areas. PocDoc is a great tool to use for events such as concerts, sporting events, large gathers because a medical mobile clinic can be deployed for community health monitoring, prevention, and care. We have routing algorithms to determine which is the best place to go for most impact, events happening in areas, and on demand at home care and what specialized mobile clinics to deploy (a unique feature our competitors do not have). PocDoc is different than other competitors on the market because 1) finetuned an LLM that was trained on medical diagnosis data 2)Logistics handling for figuring out where to dispatch mobile clinics 3) Patient has options of on-demand healthcare coming to house, going to nearby mobile clinic, or scheduling telehealth empowering the patient to make the choice that is best for them and fits their budget. Important to note, the goal of PocDoc is not to replace an actual medical professional diagnosis but rather to narrow down on possible diagnosis based off certain metrics to reduce triage and wait times.

Challenges we ran into

The first challenge we ran into was finding an LLM on huggingFace that was 1)did not require author permission (open source without having to wait for approval) 2)was trained on medical diagnostic data. Once we found a sufficient LLM, we had to figure out how to run it without a cloud host. We used an api to connect the endpoint that was hosted on HuggingFace dedicated servers. After that, we had to combine the LLM and OpenAI NLP abilities (LLM dont have dialog capabilities like NLP does) together to create a AI Health assisted chatbox that has the diagnosting prediction abilities of the LLM and the NLP and memory capabilities of OpenAI. The next challenge we ran into was HuggingFace crashing as well as Intel developer cloud. The next challenge we ran into was combining our frontend design with the backend capabilities. In retrospect, we could have planned this out better. It was hard to figure out how to connect python powered functions with typescript/javascript frontend. We decided to pivot to switch to javascript powered functions to combine with next.js frontend, however our inexperience got the best of us as figuring out next.js was becoming too time-consuming. Another challenge we faces was determining the most significant factors when it comes to which areas are the most in need. In the end we ran correlation tests (running actual matrixes test, big maths) to determine where our impact could be the most felt. Another challenge we faced was mapping a subset of real-time global data and trying to build for future scalability. We overcame this challenge by creating filters to determine units based on hospitals and patients based on units. Because our pitch is contracting out medical resources, we needed to figure distribution of medical resources to ensure taking away medical professional in one area was not negatively impacting another.