RAA — The Self-Improving AI Workspace

Inspiration

Neurodivergent people often have to work across tools that are visually dense, noisy, and fragmented. We wanted a calmer, more predictable workspace that reduces cognitive load without taking control away from the person using it.

What it does

RAA is a local-first AI workspace prototype that brings chat, deep research, documents, persistent memory, and multi-agent collaboration into one accessible environment:

• single-pane layout with collapsible sidebars and clear visual hierarchy
• keyboard-first navigation and an ARIA-compliant interface
• high-contrast support and a minimal-cognitive-load design system
• persistent memory so users do not have to repeatedly re-explain context
• Cowork Mode, which breaks a goal into specialist agent tasks
• local-first operation, so cloud services are optional

For this hackathon, we see RAA as a foundation for neurodiversity-aware productivity: a workspace that can adapt pacing, presentation, reminders, and assistance to the user's preferences.

How we built it

We built a full-stack prototype with a React/TypeScript frontend, Python backend, Docker support, persistent memory, an agent runtime, and MCP integrations. Agents can suggest, summarize, and coordinate work while the person decides what to accept.

Challenges and learning

The hardest part was making powerful automation feel predictable. We learned that accessibility is not only a color-contrast checklist; it is also about reducing surprise, preserving context, exposing clear next steps, and letting users choose the level of assistance.

What's next

We would validate the interaction model with neurodivergent users and add configurable sensory settings, focus modes, text-to-speech and speech-to-text, task chunking, and personalized pacing. This is a research/demo prototype, not a medical or clinical tool.

Built With

• fastapi