Aller au contenu

Quran Gallery Journal des modifications

Journal des Modifications

Suivez chaque mise à jour de la suite Quran Gallery.

25 entrées9 fonctionnalités11 améliorations3 correctifsdepuis novembre 2025

v2 · févr. 2026

Fonctionnalité v2

Quran Gallery Links in AI Responses

Our model now embeds direct links to the Quran Gallery reader when referencing Quranic verses and Islamic content.

  • AI responses include clickable links to verses in the Quran reader
  • Links open in a new tab for seamless browsing
  • Better integration between the chatbot and the main Quran Gallery platform
Amélioration v2

Multi-Turn Conversation Context

Upgraded our model to maintain full conversation history for more coherent, context-aware follow-up answers.

  • Model now processes full multi-turn message history
  • Improved contextual understanding across long conversations
  • More natural, flowing dialogue with accurate follow-ups
Fonctionnalité v2

Admin User Categorization

Admin panel now features categorized tabs for better user management and oversight.

  • Added tabbed view to categorize users in admin panel
  • Easier filtering and management of user accounts
  • Improved admin workflow for monitoring platform usage
Amélioration v2

Scholarly Study Room Chat Design

Complete chat UI redesign inspired by a scholarly study room aesthetic with warm, inviting tones.

  • New scholarly study room visual theme for the chat interface
  • Improved typography with the canonical .qg-* serif heading scale
  • Warm color palette with grain texture overlays
  • Enhanced readability and visual hierarchy for messages
Correction v2

Markdown Rendering Fix

Fixed escaped newlines in model responses so markdown renders correctly with proper formatting.

  • Normalized escaped newlines in model outputs
  • Proper markdown rendering for lists, headings, and paragraphs
  • Consistent formatting across all generated content
Amélioration v2

Dynamic Sitemap

Replaced static sitemap with a dynamic server route for better SEO and automatic updates.

  • Dynamic sitemap generation via server route
  • Automatic updates when new pages are added
  • Proper XML formatting with lastmod and priority
Amélioration v2

Islamic Editorial Homepage

Landing page redesigned with an elegant Islamic editorial aesthetic featuring scroll-triggered animations.

  • Editorial layout with numbered trust pillars
  • Scroll-triggered reveal animations throughout
  • Interactive chat preview with perspective depth effect
  • Full dark mode support with refined color tokens
Amélioration v2

Production Readiness & SEO

Comprehensive SEO optimization, meta tags, and structured data for the v2 platform.

  • Full meta tag coverage with Open Graph and Twitter cards
  • Structured data for search engine discoverability
  • Performance optimizations for edge deployment
  • Prerendered static pages where applicable
Fonctionnalité v2

Admin Dashboard & Streaming Overhaul

Major admin dashboard improvements and optimistic response streaming for a snappier chat experience.

  • Redesigned admin dashboard with detailed stats
  • Optimistic response streaming for faster feedback
  • Input length limits to prevent abuse
  • Better conversation and message management tools
Amélioration v2

Model Fine-Tuning v2 — Islamic Scholar Persona

Second-generation model tuning with deeper Islamic scholarship training data and stronger safety guardrails.

  • Custom-tuned model with deep Islamic scholarship training data
  • Enforced theological accuracy for Khatm-e-Nubuwat and core aqeedah
  • Multi-language response adaptation (Arabic, Urdu, English auto-detection)
  • Structured JSON output format for consistent answer rendering
  • Built-in jailbreak and prompt injection resistance
Fonctionnalité v2

Dual AI Engine with Automatic Failover

Integrated dual AI engines with transparent failover so users never experience downtime.

  • Primary inference engine with extended thinking capabilities
  • Secondary fallback engine for high-availability resilience
  • Integrated search grounding for real-time knowledge retrieval
  • Transparent failover — users never see downtime
  • 30-second timeout with graceful degradation
Fonctionnalité v2

Real-Time Streaming Engine

Purpose-built streaming engine for chunk-by-chunk response delivery with partial JSON parsing.

  • Server-Sent Events (SSE) for chunk-by-chunk response delivery
  • Partial JSON parsing during stream for instant answer rendering
  • Multi-turn context injection (6-message sliding window)
  • Client-side regex extraction for mid-stream answer display
  • Escaped newline normalization for clean markdown output
Correction v2

Rate Limiting & Security Hardening

Added sliding window rate limiting and admin authentication guards to protect the platform.

  • Sliding window rate limiter per user
  • Rate limits on chat, regenerate, read, write, and delete actions
  • Admin route authentication guards to prevent unauthorized access
  • Stream endpoint rate limiting for abuse prevention
  • Sliding window counter algorithm with per-user KV timestamps
  • Automatic TTL expiration to prevent stale rate data
  • Retry-After header calculation for client-side backoff
Amélioration v2

Shared Database Architecture

Consolidated the database layer into a shared PostgreSQL instance with per-app schemas and strict data integrity guarantees.

  • PostgreSQL as the single source of truth across all apps
  • Drizzle ORM with per-app schemas in the shared `@qg/db` package
  • Soft-delete pattern across all user-facing tables
  • Strategic indexing on userId, timestamps, and version uniqueness
  • Cascading foreign keys for data integrity
Fonctionnalité v2

Message Versioning & Regeneration Engine

Non-destructive version history system for regenerated responses with instant version switching.

  • Non-destructive version history for every regenerated response
  • Active version pointer for instant version switching
  • Optimized batch loading to prevent N+1 query overhead
  • Version numbering with unique constraint enforcement
  • Full context replay on regeneration for consistent answers
Annonce v2

v2 Platform Rewrite

Complete platform rewrite with SvelteKit — faster, lighter, and built for the edge. Everything rebuilt from the ground up.

  • Full rewrite with SvelteKit 2 and Svelte 5
  • Edge-first architecture for global low-latency responses
  • New authentication system with Google OAuth
  • Real-time streaming for instant model responses
  • Redesigned chat experience with modern UI
  • Per-request database and auth binding lifecycle for Worker compatibility
  • Valibot schema validation at all system boundaries

v1 · nov. 2025 – janv. 2026

Fonctionnalité v1

Message Version History

Users can now browse previous versions of regenerated messages, making it easy to compare answers.

  • Version tracking for regenerated AI responses
  • Browse and compare previous message versions
  • Persistent version history across sessions
Amélioration v1

Auto-Login with Google

Streamlined the sign-in experience with automatic Google OAuth login for returning users.

  • One-click Google sign-in for returning users
  • Reduced friction in the authentication flow
  • Seamless session persistence across visits
Fonctionnalité v1

Admin Dashboard

Built an admin dashboard for monitoring users, conversations, and platform health.

  • User and conversation overview for admins
  • Message and usage statistics
  • Ability to manage conversations and user accounts
Correction v1

Rate Limiting & Security Prompt

Added rate limiting to prevent abuse and hardened the system prompt for safer model responses.

  • Per-user rate limiting on API endpoints
  • Hardened system prompt to prevent prompt injection
  • Improved input sanitization for user messages
Amélioration v1

Model Fine-Tuning v1

Replaced the general-purpose model with our own fine-tuned Islamic knowledge model for more accurate, scholarly answers.

  • Trained on curated dataset of classical and contemporary Islamic scholarship
  • Specialized in Quranic tafsir, hadith verification, and fiqh rulings
  • Calibrated response confidence to distinguish strong consensus from scholarly debate
  • Multi-source cross-referencing for hadith authenticity grading
  • Iterative evaluation against 150+ theological edge cases
Amélioration v1

Prompt Engineering & Knowledge Calibration

Extensive prompt engineering and knowledge calibration to ensure accurate, complete, and well-sourced Islamic scholarship responses.

  • Curated Islamic knowledge corpus covering Quran, Hadith, Fiqh, and Seerah
  • Response completeness enforcement (e.g., all five pillars, not just one)
  • Citation accuracy tuning for Surah/Ayah references
  • Anti-hallucination safeguards for sensitive theological topics
  • Iterative prompt testing across 200+ sample queries
Amélioration v1

Dark Mode & UI Refresh

Added full dark mode support and refreshed the chat interface for better readability.

  • System-aware dark mode with manual toggle
  • Refreshed chat bubble design and spacing
  • Improved mobile responsiveness
Fonctionnalité v1

Production Launch

Deployed the platform to production as a set of Node.js services fronted by a reverse proxy.

  • Node.js services built with `@sveltejs/adapter-node`, managed by pm2
  • Caddy reverse proxy routing path prefixes to per-app processes
  • Per-request session verification against the accounts SSO hub
  • Environment validation on cold start for fail-fast behavior
Annonce v1

Initial Launch

Quran Gallery chatbot goes live — an AI-powered Islamic Q&A assistant built to help users explore the Quran and Islamic scholarship.

  • AI-powered Islamic Q&A with scholarly knowledge base
  • Real-time streaming responses
  • Conversation history with multi-session support
  • Secure authentication with Google OAuth
  • Mobile-friendly responsive design