Designing AI Productivity Assistants for Neurodiverse Professionals

This project explores a new, inclusive approach to designing AI assistants. The goal is to create tools that act as "cognitive co-regulators" for neurodiverse professionals (e.g., individuals with ADHD, ASD, dyslexia).

Instead of generic, one-size-fits-all solutions, this research focuses on building adaptive, real-time support using a "no-code" frontend combined with a powerful "high-code" AI backend. This interactive report walks through the foundational principles, the target users, the technical architecture, key features, and the evaluation plan for this innovative study.

See the Architecture

The 'Why': A Paradigm Shift

Traditional software design often fails neurodiverse users by enforcing a single, "neurotypical" way of working. This project proposes a shift from static, universal design to dynamic, pluralistic design that adapts to the individual.

OLD: Universal Design

This "one-size-fits-all" model assumes all users benefit from the same interface and workflow.

  • Fails to account for cognitive diversity.
  • Often results in "one-size-fits-none."
  • Places high cognitive load on users to adapt.
  • Ignores sensory and emotional context.

NEW: Neurological Plurality

This model embraces diversity by designing for dynamic adaptation and individual cognitive patterns.

  • Integrates neurodivergent perspectives from the start.
  • Prioritizes reduced cognitive load and flexible use.
  • Uses AI to adapt content, cues, and schedules.
  • Shifts from static reminders to dynamic, real-time support.

The 'Who': Target Professionals

The assistant is designed for professionals with diverse cognitive profiles. Understanding their specific challenges is key to designing features that offer genuine support.

ADHD

Common Challenges:

  • Task initiation and prioritization
  • Time management ("time blindness")
  • Sustaining focus and attention
  • Impulse control

Autism Spectrum (ASD)

Common Challenges:

  • Executive functioning
  • Transitions between tasks
  • Sensory overload from notifications
  • Interpreting "unwritten" social/work rules

Dyslexia / Dyspraxia

Common Challenges:

  • Processing large blocks of text
  • Organizing written thoughts
  • Working memory overload
  • Fine motor control (for Dyspraxia)

The 'How': Hybrid (No-Code + AI) Architecture

This project uses a "hybrid" model. A simple, flexible "No-Code" frontend allows for rapid prototyping and easy user interface changes. This connects to a powerful "High-Code" AI backend that handles complex logic, personalization, and data processing.

Figure 1: System Architecture Diagram

Frontend (No-Code)

Bubble.io
→

Middleware (Automation)

Airtable
Zapier / Make
↔

Backend (Python/LLM)

LangChain
RAG Model

Click on the boxes above to learn more about each component.

The 'What': Inclusive Feature Design

Based on the needs of neurodiverse users, the assistant will focus on features that reduce cognitive load and support executive functions.

1. AI Task Breakdown

User enters a large, vague task (e.g., "write report"). The AI automatically breaks it down into small, concrete, and manageable steps.

2. Multimodal Cues

Allows users to choose how they get reminders. Options include simple text, visual timers, gentle sounds, or "quiet" notifications that don't break focus.

3. Reduced Cognitive Load

The interface itself is minimal and customizable. Users can hide buttons, change text size, and use a "focus mode" to see only their current task.

4. Adaptive Schedule Nudges

Instead of rigid alarms, the AI monitors task progress and gently "nudges" the user, (e.g., "This task seems to be taking a while. Would you like to take a break?").

5. Custom Content Presentation

The AI can summarize long emails, pull out key action items, or re-format text with shorter lines and more white space to aid in reading.

6. Conversational Memory

The user doesn't have to repeat context. The AI remembers past conversations and documents, making interactions feel more like talking to a real, helpful assistant.

Prototype: ✨ AI Task Breakdown

Simulate how the AI turns a massive task into manageable steps, ideal for addressing challenges like **task initiation** and **executive dysfunction**.

Prototype: ✨ AI Research Assistant (Live Web Search)

This simulates the **RAG (Retrieval-Augmented Generation)** feature by using the **Google Search** tool. It allows the AI to access real-time, external information to answer questions, finding data that isn't in its pre-trained knowledge.

The 'Test': Evaluation & Calculation

To prove the assistant is helpful, the study will measure usability, task success, and, most importantly, the user's "perceived cognitive load." This is measured using a standard tool called the NASA-TLX.

Interactive NASA-TLX Calculator

The NASA-TLX score is a weighted average. First, the user rates the task on six scales (0-100). Then, they perform 15 "pairwise comparisons" to decide which factors were most important. This comparison count gives each factor a "Weight" (from 0 to 5). The final score is calculated: Score = Σ(Rating × Weight) / 15.

50
50
50
50
50
50

Perceived Cognitive Load (0-100)

--.--

Visuals: Sample Data & Charts

This section provides sample visuals you can generate for your research article. These charts show the kind of data you would collect to demonstrate the assistant's effectiveness.

Figure 2: Task Success Rate

Caption: A bar chart comparing the average task completion success rate for participants before and after using the AI assistant.

Figure 3: Perceived Cognitive Load

Caption: A radar chart showing the average NASA-TLX sub-scores for a complex task, comparing the "Baseline" (no assistant) to the "With AI" condition.