Skip to main content

Technology Stack

Technology Stack: Current Snapshot​

The RTS prototype combines modern web, database, and AI technologies to scaffold student inquiry, reflection, and narrative development.

Frontend​

  • Framework: React + Next.js (Pages Router), TypeScript
  • Styling: Tailwind CSS
  • UI Components: shadcn/ui + Radix UI
  • Authentication UI: Supabase Auth UI + Cookie Sessions

Backend​

  • Database: Supabase (PostgreSQL) with Row Level Security (RLS)
  • Authentication: Supabase Auth (email/password + cookie-based session)
  • API Routes: Next.js API handlers (Pages Router)
  • Deployment: Vercel (production), GitHub Codespaces (development)

AI Integration​

  • Model: Google Gemini (latest available) via @google/genai SDK
  • Output Modes:
    • Structured JSON via responseMimeType: "application/json" + responseSchema for question generation
    • Pure markdown for synthesis generation
  • "Thinking" Tokens: thinkingConfig with includeThoughts: true for surfacing intermediate computational artifacts
  • Model Flexibility: Dynamic model_override parameter for testing different models without code changes
  • AI Observability: Metadata capture at every interaction: token counts (prompt, response, "thinking", total), model string, "thought" summaries, timestamps
  • Prompt Architecture: System instructions externalized into separate .ts files, decoupled from handler logic

For details on how AI is constrained and positioned pedagogically, see A Note on AI Use.

Black Box Micro-Engagement (BBME) System (In Development)​

  • Content Management: Template-based BBME creation with course assignment
  • 4-Part Reflection Structure: Action step completion, personal reflection, relational reflection, source documentation
  • Instructor Tools: Creation, assignment, and management dashboard
  • Student Workflow: Course-scoped BBME access and submission
  • AI Synthesis: Gemini-powered reflection analysis with full metadata capture

Architectural Philosophy​

Three commitments shape every design decision in RTS.

Pedagogy-Oriented Orchestration. System design scaffolds thinking. Student reflections are structured artifacts persisted in the living ledger. Reflection trails build context across rounds: each round's AI output is grounded in the student's previous writing, which means the model has nowhere to go except deeper into what the student has already said.

Instructor-Student Ecosystem. The walled garden course structure enforces RLS boundaries. Structured assignment workflows (Movement 1, Deep Dives, BBMEs) create a complete audit trail of student work and AI interactions. Instructors see the journey, not just the product.

Questions Over Answers. Every API route is designed as a compositional aid. The system is built for recursive, inquiry-driven use across movements. AI generates questions and reorganizes text: the pipeline is designed to prevent it (as much as possible at the instruction level) from introducing new content. (I obviously can’t see the hidden system instructions the model is following).