Inspiration

If you have ADHD—and specifically Rejection Sensitive Dysphoria (RSD)—you know how exhausting a single day can be. A two-word text from a manager ("sounds good") or a friend canceling plans doesn't just feel like a minor bump; it can trigger a full-blown "doom spiral." Your brain actively rewrites the neutral event into a catastrophic narrative: They hate me, I'm bad at my job, everyone is angry with me.

We realized that standard journaling apps or simple AI chatbots fail in these moments. When you are caught in a spiral, you don't need a blank page, and you don't need a robot telling you to "just take a deep breath." You need an active intervention. We wanted to build a tool that feels like an empathetic, ruthlessly logical friend who can interrupt your spiral in real-time and help you untangle your anxious thoughts before they ruin your weekend.

What it does

SocialRewind is an AI-powered debriefing space designed specifically for the neurodivergent experience.

  • The Doom Spiral Detector: It acts as a digital grounding technique. As you type out your anxieties, the interface actively reads your phrasing. When you start using catastrophizing language, the UI subtly reacts and pulses, bringing you back to the present moment and making you aware of the spiral as it happens.
  • The ADHD Lens: It acts as an impartial mediator. You do a chaotic "brain dump" into the app, and the AI gently separates the undeniable facts of what happened from the anxious stories your brain is attaching to those facts.
  • Perspective Flip: This is cognitive restructuring made visual. It explicitly contrasts your "Brain Version" ("They are furious with me") directly against a highly logical "Reality Version" ("They were clearly just in a rush in traffic").
  • The Wins Wall: It’s scientifically proven that the ADHD brain fixates on failures and ignores successes. SocialRewind forces balance by actively maintaining a Wall of Wins to rebuild confidence over time.

How we built it

We knew the technology had to disappear into the background to keep the cognitive load as low as possible for someone in an active state of panic.

We built the Frontend to be a calm, dark-mode safe space using Next.js and Framer Motion. The interactions needed to feel fluid, seamless, and immediate. We spent hours perfecting the "magic moment" where the text box reacts to your typing, ensuring it felt like a supportive nudge rather than a glaring red error message.

For the Backend, we needed a system capable of handling complex reasoning without making the user wait. We used FastAPI and structured the application using the Model Context Protocol (MCP) to seamlessly route information between the user, their historical patterns, and the AI.

The core of the experience relies entirely on profound empathy. To achieve this, we turned to DigitalOcean's Gradient AI Workspace. Generic LLMs often give robotic, dismissive advice to neurodivergent users. By deploying a tailored Llama 3 model on DigitalOcean's serverless endpoints, we had the explicit control needed to tune the AI's personality. We trained it to respond not as a chatbot, but as an expert, warm, and highly specific ADHD coach.

Challenges we ran into

  1. Teaching AI to "Get It": The hardest part wasn't the code; it was the psychology. Prompting the Llama 3 model to respond with the exact tone needed—empathetic, clinical, validating, yet firm enough to challenge distortions—took massive refinement. It had to fundamentally understand the mechanics of executive dysfunction and "masking."
  2. Designing for Sensory Overload: When someone is spiraling, their sensory bandwidth is zero. We had to constantly pull back on our UI designs. We learned to make the real-time Doom Spiral warnings subtle enough that they didn't induce more anxiety or sensory overwhelm.
  3. Frictionless Entry: We had to figure out how to get a user from "panicking" to "debriefing" in fewer than two clicks. If an ADHD user has to navigate three menus to log a journal entry, they simply won't do it.

Accomplishments that we're proud of

We are incredibly proud that we built a piece of software that actually makes people feel seen. We've created an interface that doesn't just passively record data, but actively participates in emotional regulation. Moving the project from a rough concept of "AI journaling" into a deeply specialized, therapeutic tool—and successfully linking that fluid frontend experience to a robust DigitalOcean AI backend—feels like a massive win for accessible mental health tech.

What we learned

We learned that building software for neurodivergence isn't just about high-contrast fonts or screen readers; it's about cognitive accessibility. It's about designing architecture that assumes the user might be operating at 10% executive function and 100% panic. We also learned how powerful serverless AI deployments (like Gradient) can be in offloading the heavy lifting of natural language processing so we could focus entirely on the human experience.

What's next for SocialRewind

  • Proactive Wearable Hooks: Integrating with smartwatch HealthKit APIs to monitor heart-rate variability (HRV). The app could proactively ping you and ask if you need a debrief just seconds before you even realize you're entering a spiral.
  • Full Voice Integration: Connecting deep audio transcription so users don't even have to type—they can just talk out loud while walking their dog and watch the app deconstruct their spiral live.
  • Therapist Export: Creating a secure bridge so users can export their RSD Heatmaps and interaction histories directly to their mental health professionals for truly data-driven therapy sessions.

Built With

Share this project:

Updates