Changelog
All notable changes to the NOBOGYAN Learning System
v1.3.0
June 19, 2026
Site-Wide Audit & UX Overhaul
Added
- Redesigned new-path loading screen: dark constellation animation with progress ring and particle effects
- Redesigned new-path preview page: dark gradient hero header with staggered entrance animations
- Simulated incremental node appearance during path generation with countdown timer
Fixed
- Greenlet async error in hierarchical graph service (db.expire → db.refresh)
- Tutor session crash: widened current_node_id column from VARCHAR(50) to VARCHAR(500)
- Analytics back button now navigates to notebook with correct graph context
- New-path back button goes to profile-summary when no active graph exists
- Logo link changed from href="#" to href="/"
- Removed duplicate "tap any topic" hint from graph component
Changed
- Removed fabricated social proof: fake testimonials, trust bar, metrics, and 2,547+ waitlist counter
- Removed fake 4.9-star / 2,000-review aggregateRating from structured data
- Replaced all Sparkles icons with contextual alternatives (Target, Zap, BookOpen, etc.)
- Rewrote landing page copy: Hero, FinalCTA, AI Tutor, FeatureSpotlight, TheSwarm, Footer
- Updated social links in footer and structured data to real accounts (X, YouTube, veloxstudio.tech)
- Fixed blog CTAs from external veloxstudio.tech to internal /register
- Reframed Partners page: "Partners & Tech Stack" with honest iFlytek description
- Removed "Phase 02:" / "Phase 03:" labels from onboarding pages
- Deleted 6 dead files (~3,400 lines): page-old.tsx, Metrics.tsx, Testimonials.tsx, TrustBar.tsx, LiveStats.tsx, metrics.tsx
v1.2.0
June 18, 2026
Authentication, Onboarding & Streaming
Added
- JWT authentication on all API endpoints
- Onboarding tour for first-time notebook users with panel interaction
- Streamed quiz generation for 3-4x faster perceived response
- Streamed learning path generation — milestones appear in ~60s, subtopics fill progressively
Fixed
- Learning path now starts from the first milestone correctly
- Onboarding tour text updated to match actual project features
- Stale data display and streaming UX issues in path preview
- Docker build, Netlify routing, shadowed import, compose deprecation
Changed
- Prioritized first milestone subtopic generation for faster initial load
v1.1.0
June 16-17, 2026
Kimi k2.6 Migration & Stability Fixes
Added
- Two-pass lazy learning-path generation for better performance
- Parallel subtopic materialization
Fixed
- Quiz results 500 error: select latest attempt instead of scalar_one_or_none
- Quiz results 422 error: pass student_id as query param
- Mermaid syntax-error crashes in AI tutor: guard partial/streaming diagrams
- Code agent truncation and parallelized grading
- Faithfulness fail-closed + notes partial-credit gate logic
- Milestone progression, navigation, gate tracking, and async remediation (E2E fixes)
- Silent template fallbacks in LLM provider
Changed
- Migrated to Kimi k2.6 as production LLM provider with reasoning toggle
- Streamed resources to frontend in real time
v1.0.0
May 21, 2026
Major ReleaseInitial Release
Core Platform
- Multi-Agent Architecture: 15+ specialized AI agents working in coordination
- Orchestrator Pattern: Central agent for task delegation and coordination
- FastAPI Backend: High-performance async REST API with SSE streaming support
- Next.js 16 Frontend: Modern React app with App Router and Server Components
- Docker Compose Setup: Complete containerized development environment
- Firebase waitlist integration with performance optimizations
- SEO: favicon, meta tags, Open Graph, Twitter cards, JSON-LD structured data, robots.txt, sitemap
Feature 1: Conversational Learner Profiling
- Natural language dialogue-based profile extraction
- Six-dimensional learner model (Knowledge, Cognitive Style, Weak Points, Goals, Pace, Preferences)
- Embedding-based gap detection using cosine similarity
- Dynamic profile updates with confidence-weighted averaging
- Chat-based onboarding UI
Feature 2: Multi-Agent Resource Generation
- Scholar Agent: Lecture notes and reading materials (Markdown)
- Mapper Agent: Interactive visual mind maps (JSON → SVG)
- Sage Agent: Adaptive assessments with difficulty scoring
- Director Agent: Video scripts with TTS narration
- Architect Agent: Programming exercises with Judge0 sandbox integration
- Parallel async generation with faithfulness verification
Feature 3: Adaptive Learning Path Planning
- A* search algorithm over knowledge graph
- 500+ cloud computing topics indexed
- Dynamic path replanning based on quiz performance
- Milestone-based progression with gates
- Content-based + collaborative filtering recommendations
Feature 4: Real-Time AI Tutoring
- Streaming SSE responses with rolling context window (8,000 tokens)
- Multimodal input/output (text, voice, image, diagrams)
- RAG grounding with Weaviate vector database
- Profile-aware responses with auto-generated Mermaid diagrams
Feature 5: Learning Assessment and Analytics
- Adaptive quiz difficulty based on mastery level
- LLM-based short answer grading
- Code execution via Judge0 sandbox
- Gate-based milestone unlocking
- Comprehensive analytics dashboard with LLM-powered insights
Future Roadmap
v1.4.0 — Planned
- Teacher dashboard with class-level analytics
- Video generation from Director agent scripts
- Console log cleanup across frontend
- Self-hosted fonts (replace external Fontshare CDN)
- CORS lockdown for production deployment
v2.0.0 — Future
- Multi-language support (Chinese, English, more)
- Mobile app (React Native)
- Offline mode for learning without internet
- LMS integration (Moodle, Canvas)
- Extended knowledge graphs beyond cloud computing
For questions about the changelog: theveloxstudio@gmail.com