I Went Viral!: Building An AI System For Sensory Processing

Using AI for good to help neurodivergent people just like me.

Photo by Pixabay from Pexels: https://www.pexels.com/photo/grayscale-photography-of-bridge-262367/

Firstly…

This project began as a response to a personal challenge as an autistic women and an ADHDer. The last couple of days have been so amazing. I shared a TikTok talking about a ML model that helps prevents and predicts sensory overload in autistics and ADHDers and it went viral on LinkedIn! I’m so overwhelmed with the positive comments, the community being built around it, the optimism… thank you all so, so much!

The project

Building an AI system for the neurodivergent community is so important to me especially as a neurodivergent woman. We have so many tools that we have to use to manage our symptoms but many are not made with accessibility and neurodivergent brains in mind. For example, many women and non-binaries with ADHD tend to have different experiences in the lead up to and during their monthly cycles. It’s an experience many people can relate to but not something considered when tools and support is being discussed.

This machine learning project I’ve built is the beginning of something that could make such an impact and I’m excited to build it, grow it and share it with the world especially our neurodivergent community.

The problem

Sensory overload. Some of you may have heard that term. Some may interpret it as simply being overwhelmed but for us, as neurodivergent individuals, particularly autistic people and those with ADHD like me.

Sensory overload is defined as:

“an occurance when an individual's brain becomes overwhelmed by too much sensory input, making it difficult to process the information effectively. This can involve overstimulation from one or more senses, such as sight, sound, touch, taste, or smell. Common triggers include noisy or crowded environments, flashing lights, or strong odors. Symptoms can vary but often include anxiety, irritability, difficulty focusing, or even shutting down to cope with the overwhelming stimuli.”

Sensory overload is something I’ve experienced a lot in my life; from public places to within the comfort of my own home. What inspired to build this project was my partner, who is also autistic. When she experiences sensory overload, it presents differently to how I present it. For her, she may get frustrated, start to shut down vocally and that’s a sign that wherever we are, we need to exit that area quickly. For me, in heightened moments of stress, my sensory overload can present into bursts of emotion, it may be seen as angry outbursts to others. I flap my hands sometimes trying to calm down and I cry.. a lot! I also experience an increase in body temperature and sweating.

Sensory overload is a daily reality for us. From working in offices with incredibly bright lights that feel overwhelming to needing to wear headphones to protect us from the immense amount of noise we navigate in public spaces.

We use many tools to help us navigate sensory overload. Some may wear sunglasses, others use noise cancelling headphones (like myself!) and for many, avoiding certain environments altogther.

Altogether, these are helpful tools and strategies, they’re focused on being reactive rather than predictive. They don’t account for individual experiences and differences such as the additional experience of women or people with a uterus.

AI gives a great opportunity to combine environmental monitoring with machine learning. We can create systems that understand us as neurodivergent people, that can predict and adapt to individual sensory needs.

The technical

My solution uses a hybrid AI model that combines general understanding with personal adaptation. I wanted to create something that could truly understand and adapt to individual sensory experiences: like a ‘sensory’ weather app.

I started by using synthetic data modelling common sensory patterns which include data points such as:

• Environmental Monitoring

  • Light levels

  • Sound intensity

  • Movement patterns

  • Flourescent vs natural light

• Movement Analysis

  • Visual motion tracking

  • Space dynamics

  • Crowd density

  • Pattern recognition

• Environmental Factors

  • Temperature variations

  • Humidity levels

  • Air quality

I chose these points because from research, as well as past personal sensory overload experience, these factors seemed like a great place to start as a foundation. For it to really do what it does best, personalisation is where the innovation would be impactful as it’ll learn the patterns of the individual.

As I further develop this, my aim is to expand its capabilities to do things such as:

1. Enhanced Personalisation:

• Multiple context support

• Predictive modeling

2. Integration Features:

• Wearable device support

• Alert systems

• Automated adjustments

3. Community Aspects:

• Anonymous pattern sharing

• Trend analysis

• Community insights

I want this AI system to protect the user, to be ethical, to be inclusive and to be helpful to us and this is just the beginning!

The goal is to create technology that truly understands and supports neurodivergent experiences, making spaces more accessible for everyone. I’m constantly learning and adapting, just like our AI system!

The challenges

A question I was asked was why I chose to use synthetic data in the model. There was a multitude of reasons. The top reason was I’m still trying to figure out how to source data for this ethically. It would allow me to create a prototype faster while I research into ethical data collection practises.

As it further develops, the base model will get personalised through 2-6 weeks initial calibration period, a minimum of 100 data points for a basic personalisation. I believe that 500 data points would be an ideal point to build to in order to gain good accuracy as well as getting user feedback and response tracking.

Each individual is different. What is overwhelming for me may not be overwhelming for another. This is why I opted for creating a base Machine Learning model that can be used straight away so I can gradually move to more personalised predictions that’ll continually learn and adapt thresholds.

Community feedback will be essential for its growth. Although I’m depending a lot on my neurodivergent experience, the feedback from the community is what will make this even greater and hugely beneficial for the community.

The goals

My immediate goals is to continue to build on this project for it to become a tool or a feature for neurodivergent people. I plan to continue to share how it evolves and continue to build community around it. This is something being built for neurodivergent people by a neurodivergent woman but I need the input of my community to make it something that we all need it to be: a system that understand us.

I’m learning along the way while building it publicly; sharing the good, the bumps and the confusing. It’s definitely challenging this balancing an exciting project while working full time as a Data Engineer and studying full time for my Bachelor of Science degree. If I could put all my time to this, I would but at this moment in time, I’m unable to.

I’m excited to keep growing this and sharing the process. This feels like the beginning of my own research lab with inclusivity, ethics and our neurodivergent brains in mind.

Let me know your thoughts, leave any comments you may have and if there’s more you’d like to learn about this specific project or the AI system I’m building, let me know!

With love and neurospice,

dbmih xo