Inspiration

Exquisicare was inspired by the undeniable growth of both an opioid epidemic within the United States and a unique opportunity presented by the equally unrelenting growth of social media use and screen time. The numbers don’t lie - the age-adjusted rate of opioid related overdose deaths rose 31% from 2019 to 2020, with almost a tenth of the total drug-related deaths ever recorded being realised in the US in 2020 at around 91,000. Equally, screen time has grown to an average of over 7 hours a day. With the dopamine rush from a DM often coinciding with negative connotations, Exquisicare wants to leverage it for the better by offering a free to use, AI-backed, responsive and HIPAA-compliant chatbot to help advise and guide those worried about the risk of addiction in the right direction.

What it does

The Exquisicare bot responds to queries through Instagram and then uses continuously improving intent recognition to understand the desire of the user. When an understanding is achieved, Exquisicare will prompt for more information from the user if they are asking about likelihoods, concerns, facts or any other quantitative opioid-related information. After making a disclaimer and with the consent of the user, high level demographic information is taken and input into a classification model trained in vast amounts of CDC data to better inform the user of their real risks, and then reassure them by offering them assistance via a medical professional who can help them stay clear of addiction if at risk.

Eventually, Exquisicare will be able to also securely record addict patient data and use it to improve its models.

How I built it

Exquisicare was built using a python backend server leveraging Google Firebase for non relational/unstructured data storage, including intents and chat metadata, as well as a number of machine learning libraries and NLP modules from nltk in tandem with an open Source Instagram API for making rest calls. Data was sourced from the CDC, and stored encrypted on our firebase storage instance.

Challenges I ran into

It was a real challenge to source sufficient structured and available opioid-addict patient data that included features that could be tied in with information that most people would feel comfortable sharing - the majority of features pertain to existing prescription rates in the area of study and are not patient-level, which makes it hard to make the feedback to the patient specific and tailored to them. Additionally, the full extent to which we needed to abide by technical regulations in order to be HIPAA compliant was certainly cumbersome and a lot to manage for just a few days - but we feel we covered everywhere we could given the time constraints.

Accomplishments that I'm proud of

I’m really happy that I managed to get this working by myself in just a day or two. It was a real push but hugely rewarding and has renergised me to spend more time building the platform out given how widely used Instagram is and hence the reach this could have. I’m also really happy that I was able to work with an NLP + classification model in parallel and have them both hooked up to firebase, giving the platform a scalable pipeline for its future rollout.

What I learned

I refreshed a lot of my memory on text vectorisation methods and classification model evaluation. I would have liked to have done more feature analysis and spend more time making the classification multivariate instead of univariate (time constraint) for the proof of concept. I also learned a significant amount about the opioid crisis in the US and have fully appreciated the magnitude of it given the time and effort so many people are putting into trying to find a solution to it.

What's next for Exquisicare

Next up, we’d like to review and improve our model to make it highly patient-centric, make ourselves 100% compliant in all areas that may be outstanding and build our team. Then we’d like to trial a social media marketing strategy amongst young adults and create a trendy brand image that changes the boring or negative stigma around communicating with large organisations through social media that are usually labelled as out of touch.

Built With

Share this project:

Updates