Semantic Restructuring Canvas

AI cognitive triage for sleep-deprived engineering students in 3AM distress. Converts chaotic brain-dumps into React Flow graphs tagging CBT distortions with Socratic reframes;

Comment

Semantic Restructuring Canvas (SRC)

Inspiration

The idea came from a real gap in high-pressure academic environments like IITs. Students experiencing distress late at night are unlikely to seek formal help but are often willing to write their thoughts somewhere. SRC targets this exact moment — bridging the gap between midnight distress and professional care.

What It Does

SRC is a non-conversational AI cognitive triage tool.

  • Users input raw thoughts (unstructured text)
  • The system identifies cognitive distortions (e.g., catastrophizing, black-and-white thinking)
  • Instead of chat, it returns a visual node graph, forcing active thinking and restructuring

Key Features:

  • Visual Thinking: Graph-based restructuring instead of passive chat
  • Safety First: Detects crisis signals (even in Hinglish/slang) and locks UI with referral guidance
  • Privacy-Focused: No login, no storage — everything is ephemeral

How It’s Built

  • Architecture: Next.js 14 (monolith), deployed on edge for fast response
  • AI Layer: Claude with structured tool-calling → ensures safe, deterministic outputs
  • Frontend: React Flow (graph UI), Zustand (state), Tailwind CSS

  • Safety Layer:

    • Schema validation (Zod)
    • Rate limiting (Redis)

Key Challenges

  • Preventing AI from giving clinical advice or diagnoses
  • Avoiding fake empathy (no chatbot behavior)
  • Detecting culturally specific distress signals (e.g., Hinglish, student slang)

What Makes It Strong

  • Zero Data Retention: Fully ephemeral system
  • Cultural Awareness: Understands mixed-language inputs
  • Accessibility: Designed to be usable for all users

Core Insight

In mental health systems:

A false positive (overreacting) is far better than a false negative (missing someone at risk)

What’s Next

  • Integration with university platforms (like LMS)
  • Cultural adaptation for Indian mental models
  • Offline/PWA version for accessibility

Core API Logic (Simplified Flow)

  • Input text → AI classification
  • Structured JSON output (validated)

  • Response:

    • Graph topology
    • Crisis flag

Built With

Share this project:

Updates