🚀 Moj AI - AI-Powered Slovenian Building Legislation Assistant

Version: v11.0 (AI Orchestration V7.2.1) | Last Updated: October 20, 2025

Advanced AI orchestration platform for Slovenian building permits and legislation Combining Claude Sonnet 4 (or higher), GPT-4o (or higher), and Perplexity with multimodal RAG for research-quality legal analysis

Live Application: app.mojai.xyz

🎯 What is Moj AI?

Moj AI is a production-grade, multi-tenant SaaS platform that helps users navigate complex Slovenian building legislation using advanced AI orchestration. The system intelligently combines multiple AI models, legal document databases, and real-time internet research to deliver comprehensive, source-backed answers.

Core Value Proposition

🎯 Specialized Knowledge: Multi core Slovenian building legislation PDFs (Pravilnik, TSG, Uredba, Zakon, OPN_MOM, GZ, Odlok_OPN_MOL_ID) permanently loaded
🤖 Dual-Mode AI: Choose between comprehensive research (Frontier) or fast answers (Lightning)
📚 Multimodal RAG: Upload your own PDFs, DOCX, or scanned documents for conversation-specific analysis
🌐 Real-Time Research: Internet research via Perplexity for current prices, regulations, and market data
✅ Source Attribution: Every claim backed by clickable sources (RAG documents, internet links, or general knowledge)
⚡ Transparent Processing: Real-time progress showing actual AI orchestration steps

⚡ Dual-Mode AI Orchestration

Moj AI offers two AI modes optimized for different use cases:

🚀 Frontier Mode (Research-Quality Responses)

Perfect for: Complex legal analysis, building permit applications, comprehensive research

Feature	Details
AI Model	Claude Sonnet 4 (Anthropic)
Response Length	10,000-25,000 characters
Tools Used	RAG + Perplexity + General Knowledge + Anti-Hallucination
Processing Time	60-240 seconds
Question Cost	1.0 questions
Real Example	15,277 characters, 138s, 5 sources

What you get:

Comprehensive 5-10 page reports
Detailed legal analysis with court cases
Real-time market data and prices
Step-by-step procedures
Complete source citations

⚡ Lightning Mode (Fast Answers)

Perfect for: Quick questions, simple queries, cost-effective responses

Feature	Details
AI Model	GPT-4.1 April 2025 (OpenAI)
Response Length	2,000-3,000 characters
Tools Used	RAG + General Knowledge (Perplexity skipped for cost savings)
Processing Time	15-30 seconds
Question Cost	0.5 questions (half price!)
Real Example	2,929 characters, 17.1s, 3 sources

What you get:

Concise, focused answers
Essential legal information
Quick turnaround
Cost-effective (50% cheaper)

🌟 Key Features

🤖 AI Orchestration (V7.2)

✅ Dual-Mode Selection - User chooses Frontier or Lightning mode
✅ Intelligent Query Analysis - Automatically detects when RAG, internet research, or legal reasoning is needed
✅ Multi-Source Intelligence - Combines RAG documents, Perplexity research, and LLM knowledge
✅ Real-Time Progress - Live updates showing actual orchestration steps ("Searching legal documents...", "Researching current prices...")
✅ Orchestration Reports - Detailed breakdown of tools used, processing time, confidence scores
✅ Fractional Question Deduction - 0.5 questions for Lightning, 1.0 for Frontier
✅ Admin-Configurable - Master system messages, model selection, API keys managed via admin UI

📚 Dual RAG System

✅ Admin RAG - 7 core Slovenian building legislation PDFs permanently loaded for all users
✅ User RAG - Upload conversation-specific documents (PDF, DOCX, DOC, TXT, scanned PDFs with OCR)
✅ Quality Validation - Document quality scoring (0-100) prevents bad uploads
✅ Multimodal Processing - Table extraction, OCR for scanned documents, semantic chunking
✅ Source Attribution - Every RAG-sourced claim includes document name and page reference

👤 User Experience

✅ Google OAuth - Secure authentication with role-based access (user/admin)
✅ Conversation Management - Create, rename, delete, persist conversations across sessions
✅ Question Allowance - Stacking system (Premium + Bonus + Free questions)
✅ Real-Time Streaming - Server-Sent Events for live progress updates
✅ Dual Language - Slovenian/English UI with error notifications in both languages
✅ Mobile Compatible - Responsive design for all devices
✅ File Upload - Drag-and-drop document upload with quality validation

⚙️ Admin Panel

✅ AI Configuration - Configure orchestration strategies, select models (Claude/GPT-4), manage API keys
✅ RAG Management - Upload/manage 7 core legal PDFs, view extracted text, test retrieval accuracy
✅ Subscription Management - Sync Stripe products, manage subscription plans
✅ Bonus Questions - Grant additional questions to users with expiration dates
✅ System Monitoring - Real-time health checks, API status, database connections
✅ Master System Messages - Fine-tune AI behavior for Frontier and Lightning modes separately

💳 Billing & Subscriptions (Stripe)

✅ Subscription Tiers - Free (5 questions), Basic (15/month), Advanced (50/month), Max (150/month)
✅ Question Stacking - Purchased questions add to remaining balance (3 remaining + 15 purchased = 18 total)
✅ Fractional Deduction - Lightning mode costs 0.5 questions, Frontier costs 1.0 questions
✅ Bonus Questions - Admin can grant bonus questions that stack on top of purchased questions
✅ Stripe Integration - Checkout sessions, webhook handling, subscription management

🛠️ Technology Stack

Frontend

Technology	Purpose
Next.js 14	React framework with App Router
TypeScript	Type-safe development
Tailwind CSS	Utility-first styling
Shadcn/ui	Modern UI components (Radix UI)
Zustand	State management
Server-Sent Events	Real-time progress streaming

Backend

Technology	Purpose
FastAPI	High-performance async Python web framework
Python 3.11	Core programming language
SQLAlchemy	ORM for PostgreSQL
Pydantic	Data validation and settings
Alembic	Database migrations
aiofiles	Async file operations

Databases

Database	Purpose
PostgreSQL	Core data (users, conversations, messages, admin settings, subscriptions)
Weaviate	Vector database for RAG (semantic search, embeddings)
Redis	Caching and session management

AI/LLM Providers

Provider	Models	Purpose
Anthropic	Claude Sonnet 4	Frontier mode orchestrator, legal reasoning
OpenAI	GPT-4o, text-embedding-3-small	Lightning mode orchestrator, embeddings
Perplexity	sonar-pro	Real-time internet research (Frontier mode only)

Infrastructure & Services

Service	Purpose
DigitalOcean App Platform	Application hosting (backend + frontend)
DigitalOcean Managed PostgreSQL	Production database
DigitalOcean Managed Redis	Production cache
DigitalOcean Spaces	S3-compatible object storage for file uploads
Docker Compose	Local development environment
Google OAuth	Authentication
Stripe	Payment processing and subscriptions

🏗️ System Architecture

High-Level Overview

┌─────────────────────────────────────────────────────────────────┐
│                         User Interface                          │
│                    (Next.js 14 + TypeScript)                    │
│  - Chat Interface  - Admin Panel  - Subscription Management     │
└────────────────────────────┬────────────────────────────────────┘
                             │
                             ↓
┌─────────────────────────────────────────────────────────────────┐
│                      FastAPI Backend                            │
│  - REST API  - Server-Sent Events  - Google OAuth  - Stripe    │
└────────────────────────────┬────────────────────────────────────┘
                             │
                             ↓
┌─────────────────────────────────────────────────────────────────┐
│                   V7 AI Orchestration Engine                    │
│                                                                 │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐        │
│  │Query Analyzer│→ │Tool Executor │→ │Response Synth│        │
│  └──────────────┘  └──────────────┘  └──────────────┘        │
│                                                                 │
│  Tools: RAG Search | Perplexity | Legal Reasoning              │
└────────────────────────────┬────────────────────────────────────┘
                             │
                ┌────────────┼────────────┐
                ↓            ↓            ↓
         ┌──────────┐  ┌──────────┐  ┌──────────┐
         │PostgreSQL│  │ Weaviate │  │  Redis   │
         │(Core Data)│  │(Vectors) │  │ (Cache)  │
         └──────────┘  └──────────┘  └──────────┘
                             │
                ┌────────────┼────────────┐
                ↓            ↓            ↓
         ┌──────────┐  ┌──────────┐  ┌──────────┐
         │ Claude   │  │ OpenAI   │  │Perplexity│
         │Sonnet 4.5│  │GPT-4o-mini│  │sonar-pro │
         └──────────┘  └──────────┘  └──────────┘

V7 Orchestration Flow

Query Analysis - Detect if query needs RAG, internet research, or legal reasoning
Tool Selection - Choose required tools based on analysis and selected mode (Frontier/Lightning)
Tool Execution - Execute tools in parallel/sequence with context passing
Response Synthesis - Combine results with inline citations and source attribution
Quality Validation - Verify response quality and completeness
Usage Tracking - Deduct appropriate questions (0.5 or 1.0) based on mode

See: docs/architecture/ai-orchestration.md for complete details

📚 Documentation

Comprehensive documentation is available in the docs/ directory and the public documentation repository at github.com/talirezun/moj-ai-docs.

Architecture Documentation (`docs/architecture/`)

Document	Description
system-overview.md	High-level system architecture, component interactions, data flow
ai-orchestration.md	V7.2.1 dual-mode orchestration, anti-hallucination tool, response synthesis
database-schema.md	Complete database schema, relationships, indexes
multimodal-rag-architecture.md	Dual RAG system, document processing, quality validation
authentication-system.md	Google OAuth implementation, JWT tokens, role-based access
master-system-message.md	VERSION 4 AI system prompts for Frontier and Lightning modes
PROJECT_STRUCTURE.md	Codebase organization, file structure, key components

UI/UX Features (`docs/ui-ux-features/`)

Document	Description
README.md	Complete UI/UX documentation index
01_AI_ORCHESTRATION_OVERVIEW.md	Core orchestration system from user perspective
02_MODE_SELECTOR.md	Lightning vs Frontier mode selection UI
03_RAG_SYSTEM.md	Dual RAG architecture and document upload
04_QUESTION_ALLOWANCE.md	Fractional question deduction system

User Scenarios (`docs/user-scenarios/`)

Document	Description
00_USER_SCENARIOS_INDEX.md	Complete index of user scenarios
01_BUILDING_PERMITS.md	Building permit application scenarios

Public Documentation Repository

The following documentation is also available in the public repository at github.com/talirezun/moj-ai-docs:

Architecture Documentation - System design and technical architecture
UI/UX Features - User interface and experience documentation
User Scenarios - Real-world use cases and examples

💡 Use Cases & Examples

Example 1: Building Permit Application (Frontier Mode)

User Query: "I want to build a 4-unit residential building in Ljubljana's Bežigrad district. What permits do I need, what are the costs, and how long will it take?"

AI Response (10,000+ characters):

Complete list of required permits (building permit, location permit, etc.)
Step-by-step application process
Current fees and costs (from Perplexity real-time research)
Timeline estimates
Required documentation
Relevant legal articles from RAG (Zakon, Pravilnik, OPN_MOL)
Court cases and precedents
20+ clickable sources

Processing: 90 seconds | Cost: 1.0 questions

Example 2: Quick Legalization Question (Lightning Mode)

User Query: "Can I legalize a 15m² extension to my house built in 2018?"

AI Response (2,500 characters):

Yes/No answer with conditions
Key legal requirements
Simplified procedure steps
Estimated costs
4-5 relevant sources from RAG

Processing: 25 seconds | Cost: 0.5 questions

Example 3: Document-Specific Analysis (User RAG)

User Action: Uploads building inspection decision PDF

User Query: "What violations are mentioned in this document and how do I fix them?"

AI Response:

Extracts specific violations from uploaded document
Cross-references with legal requirements from Admin RAG
Provides remediation steps
Cites both uploaded document and legal framework

🔧 Development

Project Structure

moj-ai-v11/
├── backend/                    # FastAPI backend
│   ├── app/
│   │   ├── api/v1/            # API endpoints
│   │   ├── services/          # Business logic
│   │   │   ├── orchestration_v7/  # V7 orchestration engine
│   │   │   ├── rag_service_multimodal.py  # RAG system
│   │   │   ├── storage_service.py  # DigitalOcean Spaces
│   │   │   └── ...
│   │   ├── models/            # Database models
│   │   ├── schemas/           # Pydantic schemas
│   │   └── core/              # Config, auth, database
│   ├── alembic/               # Database migrations
│   └── .env.example           # Environment template
│
├── frontend/                   # Next.js frontend
│   ├── app/                   # Pages (App Router)
│   │   ├── admin/             # Admin panel pages
│   │   ├── dashboard/         # User dashboard
│   │   └── ...
│   ├── components/            # React components
│   │   ├── chat/              # Chat interface components
│   │   ├── admin/             # Admin components
│   │   └── ...
│   ├── contexts/              # React contexts (auth, usage, language)
│   ├── hooks/                 # Custom hooks (useChat, useAuth, etc.)
│   └── lib/                   # Utilities and API client
│
├── docs/                      # Documentation
│   ├── architecture/          # Architecture docs
│   ├── ui-ux-features/        # UI/UX documentation
│   ├── user-scenarios/        # User scenarios
│   ├── admin/                 # Admin guides
│   └── agent_manuals/         # Operations guides
│
├── training-files/            # Multi core legal PDFs
│   ├── Pravilnik_o_projektni_in_drugi_dokumentaciji_PISRS.pdf
│   ├── TSG-V-006_2022_razvrscanje_objektov.pdf
│   ├── Uredba_o_razvrscanju_objektov_PISRS.pdf
│   ├── Zakon_o_urejanju_prostora_ZUreP-3_PISRS.pdf
│   ├── 220528_OPN_MOM.pdf
│   ├── GZ 1.pdf
│   └── Odlok_OPN_MOL_ID.pdf
│
├── docker-compose.yml         # Local development services
├── LICENSE                    # Proprietary license
└── README.md                  # This file

Key Files

File	Purpose
`backend/app/main.py`	FastAPI application entry point
`backend/app/services/orchestration_v7/orchestration_engine_v7.py`	V7 orchestration engine
`backend/app/services/rag_service_multimodal.py`	Dual RAG system
`backend/app/services/storage_service.py`	DigitalOcean Spaces integration
`frontend/hooks/useChat.ts`	Chat logic with SSE streaming
`frontend/components/chat/ModeSelector.tsx`	Frontier/Lightning mode selector
`frontend/components/chat/QuestionAllowanceDisplay.tsx`	Question balance display

🏢 Company

Skupina Svetilnik d.o.o. Cesta 24. Junija 25 1000 Ljubljana, Slovenia

Production URL: app.mojai.xyz

📞 Contact & Support

For inquiries about:

Licensing: See LICENSE file
Partnerships: Contact via GitHub
Technical Support: Review documentation first
Private Repository Access: Contact repository owner

🙏 Acknowledgments

Built with:

Anthropic Claude - Frontier mode orchestration
OpenAI GPT-4 - Lightning mode orchestration and embeddings
Perplexity - Real-time internet research
Weaviate - Vector database
DigitalOcean - Infrastructure
Stripe - Payment processing
Next.js & FastAPI - Application frameworks

Last Updated: October 20, 2025 | Version: v7.2.1 | Status: Production Ready

Name		Name	Last commit message	Last commit date
Latest commit History 7 Commits
architecture		architecture
docs		docs
ui-ux-features		ui-ux-features
user-scenarios		user-scenarios
LICENSE		LICENSE
README.md		README.md

License

talirezun/moj-ai-docs

Folders and files

Latest commit

History

Repository files navigation