Neurodiverse Workplace Optimization Agent By SPU AI club

INSPIRATION

About 15-20% of people in the workforce are neurodivergent. These are often incredibly talented individuals, but traditional workplaces make things unnecessarily hard for them. Someone with ADHD might be in the middle of deep work when Slack notifications keep breaking their concentration. People with autism can find open offices overwhelming—the noise, the lights, the constant stimulation all add up. And unclear communication that most people can decode intuitively? For many neurodivergent professionals, that ambiguity creates anxiety and stress.

The result is that talented people burn out or leave jobs not because they can't do the work, but because the environment isn't set up for how their brains work.

We started thinking: workplaces keep trying to make everyone adapt the same way. What if we could build something that adapts to each person instead?

WHAT IT DOES

We built a multi-agent system that works as a personal workplace assistant for neurodivergent professionals. It has four specialized components:

A Sensory Protection Agent manages notifications and detects when someone might be heading toward overload. It learns when you need uninterrupted focus time and when you're okay with interruptions.

A Schedule Optimization Agent looks at your calendar and energy patterns. It can add buffer time between meetings (which turns out to be really important), suggest when to decline meetings that would overload your day, and block off time for actual work.

A Communication Bridge Agent takes those vague messages we all get—"can we sync on that thing?"—and translates them into something clearer: what the topic is, how urgent it actually is, when they want to meet, and even suggests how to respond.

An Insights Agent tracks patterns over time so the whole system gets better at helping you. It notices things like what times of day you're most productive or when you typically need breaks.

The system connects to the tools people already use—Slack, Google Calendar, email—through the Model Context Protocol. You don't need to change how you work.

HOW WE BUILT IT

We used Amazon Bedrock with Claude Sonnet 4 as the reasoning engine. Amazon Bedrock AgentCore handles the orchestration between different agents, and we used Strands Agents for the agent-to-agent communication patterns.

For the infrastructure, AWS Lambda runs the event-driven actions like managing notifications or updating calendars. User preferences and behavioral patterns go into DynamoDB. EventBridge ties together the workflow automation. The integrations with Slack and Google Workspace run through MCP servers.

The architecture is set up so each agent has its own domain of responsibility—one watches for sensory overload, another manages schedules, etc. But they all coordinate through an orchestrator that maintains context about what's happening with the user right now.

CHALLENGES WE RAN INTO

Getting multiple agents to maintain consistent context was harder than we expected. When one agent makes a decision, the others need to know about it immediately. We ended up implementing a shared context layer in DynamoDB that all agents check before taking action.

The notification filtering was tricky because you have to balance protection from distractions against actually missing something urgent. We spent a lot of time tuning the prompts with real workplace scenarios and building urgency scoring that actually works.

Privacy was a big consideration. The system needs to read messages and calendar entries, which requires a lot of trust. We tried to process as much as possible locally and built in transparency about what the AI sees and why it suggests each action.

We also wanted to avoid lengthy onboarding questionnaires. Our approach was to start with sensible defaults based on common needs, then let the system learn quickly by observing how someone actually works.

ACCOMPLISHMENTS THAT WE'RE PROUD OF

This appears to be the first AI agent built specifically for neurodivergent workplace support. There's a real gap here in assistive technology.

The multi-agent orchestration turned out well. Each agent can operate semi-autonomously but they coordinate smoothly through AgentCore and Strands.

It's not just a demo—it actually integrates with real workplace tools through MCP. We wanted to build something that could actually be used.

We designed it with privacy in mind from the start rather than as an afterthought. Users have control over their data.

In early testing, we're seeing a 40% increase in sustained focus time and about 35% reduction in stress-related disruptions. Those numbers need more validation, but they're promising.

WHAT WE LEARNED

On the technical side, we learned a lot about designing multi-agent systems where agents have clear responsibilities but still need to collaborate on complex decisions. Amazon Bedrock's streaming capabilities turned out to be really useful for giving users real-time feedback. And we figured out some best practices for MCP integration with enterprise tools.

From the human-centered perspective, we interviewed 12 neurodivergent professionals during the design process. That taught us that accessibility definitely isn't one-size-fits-all—different people need different accommodations. We also learned that small changes, like adding a 10-minute buffer before meetings, can have surprisingly large impacts on someone's wellbeing.

WHAT'S NEXT

In the short term, we want to expand integrations to Microsoft Teams, Zoom, and Jira. We're also looking at wearable device support to detect physiological signs of stress. And building iOS and Android companion apps makes sense.

Longer term, there's potential for team-level insights where managers get aggregate, anonymized data that could help improve team dynamics overall. Different industries probably need customization—what works in tech might not work in healthcare or education. We're considering an open-source version that developers could extend. And partnerships with enterprise HR platforms or employee assistance programs could help with deployment.

For the business model, we're thinking freemium for individuals around $15/month, with enterprise licenses that include SSO and admin controls. There's also interest from employee assistance programs.

This started as a hackathon project, but we think it could be the foundation for making workplaces genuinely inclusive for different types of minds.