All notable changes to LGC Concept AI are documented in this file.
This system follows Semantic Versioning (MAJOR.MINOR.PATCH).
- MAJOR → Architectural shift or breaking change
- MINOR → Feature addition or structural enhancement (non-breaking)
- PATCH → Fixes, stability improvements, automation, or internal refactoring
Date: 2026-03-13
- Progressive Web App (PWA) capability for LGC Concept AI
- Install prompt component enabling users to install the application
- Web App Manifest configuration for mobile and desktop installation
- Application icons and metadata for installable experience
- Offline-ready service worker integration
- App accessibility across devices including mobile and desktop
- Faster load performance through service worker caching
- Native-like user experience when installed as a standalone app
- Consistent launch experience via installable application shell
- Users can now install LGC Concept AI directly from the browser
- Application launches in standalone mode when installed
- Improved mobile usability for learning sessions
- No breaking API contract changes
- No backend logic changes
- Frontend infrastructure extended with PWA support
- Complete Forgot Password / Reset Password flow with secure token handling
- Password reset screen with:
- Confirm password
- Show / hide password toggle
- Strength validation feedback
- Password strength validation during user registration
- Password handling consistency across registration and reset flows
- Visual feedback for weak, medium, and strong passwords
- Safer UX by blocking weak passwords at input level
- Authentication-related pages remain focused without unnecessary distractions
- Reset password and register screens now feel consistent and intentional
- No auto-login introduced to preserve email verification model
- No breaking changes to existing users
- No new dependencies added
This release focuses on improving account recovery, password safety, and authentication user experience while keeping the system secure and predictable.
Date: 2026-02-22
- Unified multi-model fallback stack:
- Llama 3.3 70B (primary)
- Nemotron 30B (secondary)
- Gemma 27B (tertiary)
- Gemini 2.5 Flash (stable fallback)
- Provider abstraction layer (OpenRouter + Gemini)
- 429 retry handling with exponential backoff
- Per-request timeout using AbortController
- Structured request-level logging with unique requestId tracing
- Global launch transition system for controlled navigation
- Animated loading overlay with dot + progress feedback
- Rate-limit resilience across all learning modes
- Graceful degradation when providers are overloaded
- Observability of model selection and failure paths
- Frontend transition smoothness when entering Learn mode
- System-level UX coherence
- Replaced single-model dependency with unified fallback architecture
- Standardized model stack across all learning modes
- Improved API error handling (503 on total failure instead of generic 500)
- No breaking API contract changes
- No learning logic or prompt modifications
- Backend stability layer significantly strengthened
- Frontend transition logic elevated to App-level state management
Date: 2026-02-18
/explorepage to frame learning as situational/whypage to clarify system philosophyModeSwitchCTAreusable bottom navigation component- Structured entry flow via simplified Home page
- Simplified learning mode descriptions
- Removed roadmap statements and version labeling from UI
- Standardized bottom navigation across modes
- Reduced cognitive and visual noise
- Improved frontend flow coherence
- Redundant instructional text
- Top-level universal navigation clutter
- Showcase-oriented language
- No backend modifications
- No API contract changes
- No authentication changes
- Frontend structural refinement only
Date: 2026-02-17
- Separated server bootstrap into
server.jsandapp.js - Isolated MongoDB connection into configuration layer
- Extracted AI logic into
services/aiService.js - Modularized
/askroute handling - Updated CORS configuration to match Vite development server
- Cleaned dependency state after Vite optimization corruption
- Removed unnecessary SECURITY.md automation workflow
Improved separation of concerns, cleaner backend structure, stable frontend-backend communication, and simplified release flow.
Date: 2026-01-28
- Corrected SECURITY.md version tag automation workflow
Date: 2026-01-28
- Middleware-level protection improvements
- Security posture and request validation
Date: 2026-01-25
- Learn Mode
- Fast Learn Mode
- Doubt Mode
- Teach-Back Mode
- Mode-specific backend routing logic
Transitioned from single-response AI system to structured mode-based learning architecture.
Established learning as situational:
- Depth
- Speed
- Precision
- Articulation
This marked a MAJOR architectural milestone.
Date: 2026-01-18
- Stabilized email verification flow
- Improved frontend routing reliability
Date: 2026-01-08
- Email verification system
- Auth hardening
- Input validation improvements
- Security and trust layer of the application
Date: 2026-01-07
- Switched MongoDB connection to environment-based configuration
Date: 2026-01-06
- Authentication-gated usage tracking
- Activity logging per user
- Structured backend logging
Introduced identity-aware interaction layer.
Date: 2026-01-02
- Frontend wired to live backend
- Relaxed mandatory analogy enforcement
Date: 2026-01-01
- Initial AI response handling
- Basic frontend-backend communication
- Core request-response structure
Established the foundational system behavior.