Katja — Multi-Mode AI Chat Platform with Specialized LLMs

Katja is a comprehensive AI chat platform featuring 5 specialized modes: language learning with grammar correction, general chat, legal reasoning (Saul-7B), medical queries (Meditron/BioMistral), and a terminal-style console for monitoring all interactions. Built with local-first privacy using Ollama, with optional cloud fallback (OpenAI GPT-4o-mini). All data stays on your machine by default.

Features

🎓 Language Learning Mode

Multi-language support - 7 European languages (Slovene, German, Italian, Croatian, French, English, Portuguese)
Real-time grammar correction - Instant feedback with explanations in target language
Spaced repetition - SM-2 algorithm for reviewing grammar patterns
Export corrections - Save grammar corrections as JSON
Auto-clear on language switch - Fresh conversation when switching languages

💬 General Chat Mode

14+ LLM models - Choose from Llama, Qwen, GLM4, Gemma, and more
LaTeX rendering - Math and technical formulas rendered beautifully (inline $...$ or display $$...$$)
Model switching - Live dropdown selector for all available Ollama models
Structured responses - Optimized system prompts for technical and academic queries
GPT-4o-mini support - Selectable OpenAI model with automatic routing

⚖️ Legal Chat Mode

Saul-7B specialized model - Fine-tuned for English law, contracts, torts, and legal reasoning
LaTeX legal citations - Properly formatted case citations and legal principles
Precedent explanations - Detailed case law analysis
Legal doctrine support - Equity, trusts, negligence, contract law

🏥 Maritime Medical Chat Mode

Meditron & BioMistral - Medical-domain fine-tuned models
Maritime medical focus - Designed for ship's officers managing medical care without doctors
Examination guidance - Systematic patient assessment (vital signs, SAMPLE history, physical exam)
Telemedicine preparation - Organize observations for shore-based physician consultations
IMGS protocol support - International Medical Guide for Ships compliance
No diagnosis/treatment - Emphasizes observation and communication, not prescription

💻 Console Mode

Terminal-style interface - Hacker aesthetic with green monospace text
Multi-session monitoring - View conversations from all modules in one place
Real-time updates - Auto-refresh every 3 seconds
Session filtering - Filter by language learning, general, legal, or medical chat
Model tracking - See which model was used for each response
Timestamp tracking - Full conversation history with precise timestamps

Core Features

Local-first inference - Uses Ollama for privacy-preserving AI (no cloud API required)
Cloud fallback - Automatic OpenAI GPT-4o-mini fallback when selected or local models unavailable
Conversation memory - SQLite database with WAL mode for persistent chat history
Session management - Multiple conversation sessions with isolated history
Customizable prompts - Edit system prompts for each mode via settings page ⭐ NEW
CORS security - Configurable allowed origins for API access

⚠️ Important Disclaimers

Legal Chat

The Legal Chat mode is for educational and research purposes only. It is not a substitute for professional legal advice. Do not rely on responses for actual legal matters. Always consult a qualified attorney for legal questions.

Maritime Medical Chat

The Maritime Medical Chat mode is designed for ship's officers managing medical care on vessels without doctors. This tool assists with patient examination and observation to prepare for telemedicine consultations with shore-based physicians.

CRITICAL LIMITATIONS:

⚠️ NO DIAGNOSIS OR TREATMENT: Ship's medical officers are NOT authorized to diagnose conditions or initiate treatments (except immediate life-saving measures)
⚠️ PHYSICIAN AUTHORIZATION REQUIRED: All medical interventions beyond basic first aid must be authorized by a shore-based physician via radio medical advice (TMAS - Telemedical Assistance Service)
⚠️ EXAMINATION ONLY: This tool focuses on gathering complete, accurate information for remote physicians

This is NOT a substitute for professional medical advice. Always contact a shore-based physician for medical guidance.

Philosophy

Katja is a local-first, privacy-focused AI chat platform:

Privacy: All LLM inference happens locally via Ollama by default. No data sent to third parties unless cloud fallback is used.
Offline-capable: Works without internet connection (after initial model download).
Transparent: Open source backend and frontend. No telemetry.
Flexible: Choose between local models (privacy) or cloud models (performance).
Minimalist: Simple FastAPI backend + Next.js frontend.

Quick Start Guide

1. 🎓 Language Learning Mode

URL: http://localhost:3000 (default landing page)

Grammar corrections in 7 European languages with instant feedback:

Practice conversational skills in any supported language
Get real-time grammar corrections with explanations
Review corrections using spaced repetition algorithm
Export corrections as JSON for external study tools

2. 💬 General Chat Mode

URL: http://localhost:3000/chat

General-purpose AI conversations with 14+ models:

Technical discussions with LaTeX math rendering
Academic research assistance
Creative writing and brainstorming
General knowledge questions

Example queries:

"What is Bayes' theorem and how is it used in statistics?"
"Explain quantum entanglement in simple terms"
"Write a Python function to calculate Fibonacci numbers"

3. ⚖️ Legal Chat Mode

URL: http://localhost:3000/legal-chat

English law queries using Saul-7B specialized legal model:

Case law analysis and precedent research
Contract law, torts, equity, trusts
Legal doctrine explanations
Properly formatted legal citations

Example queries:

"What is the leading case in negligence?"
"Explain the doctrine of consideration in contract law"
"Is there an obligation of good faith in insurance law?"

4. 🏥 Maritime Medical Chat Mode

URL: http://localhost:3000/medical-chat

Maritime medical guidance for ship's officers using Meditron and BioMistral:

Systematic patient examination protocols (vital signs, SAMPLE history, physical exam)
Preparing observations for shore-based physician consultation
Understanding medical terminology and observations
IMGS (International Medical Guide for Ships) protocol support

Example queries:

"How should I perform a complete physical examination on a crew member with chest pain?"
"What vital signs should I collect and how do I document them for the ship's doctor consultation?"
"Guide me through assessing a patient's neurological status"

⚠️ CRITICAL: This mode does NOT provide diagnoses or treatment recommendations. It helps you examine patients and communicate findings to shore-based physicians.

5. 💻 Console Mode

URL: http://localhost:3000/console

Terminal-style monitoring interface for all LLM interactions:

View conversation history across all modules
Filter by session type (language, general, legal, medical)
Track which models were used for each response
Auto-refresh every 3 seconds
Export logs for analysis

Supported Languages

🇸🇮 Slovene
🇩🇪 German
🇮🇹 Italian
🇭🇷 Croatian
🇫🇷 French
🇬🇧 English
🇵🇹 Portuguese

Switch languages via flag buttons in the chat interface.

Prerequisites

Ollama - Install from ollama.ai for local inference

Pull models - Recommended models:

# General-purpose chat
ollama pull llama3:8b-instruct-q4_K_M

# English law (Saul 7B)
ollama pull adrienbrault/saul-instruct-v1:Q4_K_M

# Medical (Meditron)
ollama pull meditron-7b

# Biomedical (BioMistral)
ollama pull biomistral-7b

Node.js - For frontend (v18+ recommended)
Python 3.10+ - For backend
(Optional) OpenAI API Key - For cloud fallback when local models unavailable

Getting Started

Quick Start (Linux)

git clone https://github.com/SL-MAR/katja.git
cd katja
chmod +x launch-katja.sh
./launch-katja.sh

This launches:

🎓 Language Learning: http://localhost:3000
💬 General Chat: http://localhost:3000/chat
⚖️ Legal Chat: http://localhost:3000/legal-chat
🏥 Maritime Medical Chat: http://localhost:3000/medical-chat
💻 Console: http://localhost:3000/console
⚙️ Settings: http://localhost:3000/settings ⭐ NEW
Backend API: http://localhost:8000
API Docs: http://localhost:8000/docs

Manual Setup

Backend:

cd backend
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

# Copy environment template
cp .env.example .env

# Start backend
uvicorn main:app --reload

Frontend:

cd frontend
npm install
npm run dev

Configuration

Backend Configuration

The backend uses environment variables. Copy .env.example to .env and adjust:

# LLM Provider (local = Ollama, openai = fallback)
LLM_PROVIDER=local

# Ollama settings
OLLAMA_MODEL=llama3:8b-instruct-q4_K_M
OLLAMA_BASE_URL=http://localhost:11434

# Optional: OpenAI fallback (if Ollama unavailable)
# OPENAI_API_KEY=your_api_key_here

# CORS origins (comma-separated)
ALLOWED_ORIGINS=http://localhost:3000,http://localhost:3001,http://localhost:5173

Switching models: Use the dropdown in each chat mode's UI or edit OLLAMA_MODEL in .env.

Fallback behavior: If Ollama is unavailable and OPENAI_API_KEY is set, the backend will automatically fall back to OpenAI's API.

System Prompt Customization ⭐ NEW

Settings Page: http://localhost:3000/settings

Each chat mode has a customizable system prompt that defines the AI assistant's behavior:

Language Learning: Conversational teacher with grammar correction
Maritime Medical: Ship's officer examination guidance (no diagnosis/treatment)
Legal: English common law specialist
General: Technical and academic assistant

Features:

✏️ Live editor - Edit prompts in a Monaco-style textarea
💾 Browser storage - Changes saved to localStorage (persist across sessions)
🔄 Reset to defaults - Restore original prompts anytime
📋 Preview - See formatted prompt before saving
⚡ Instant effect - Changes apply on next message

Customization tips:

Adjust tone (formal/casual)
Add domain-specific knowledge
Change output format preferences
Include LaTeX notation guidance: $inline$ or $$display$$

How It Works

Architecture

┌─────────────┐      ┌──────────────┐      ┌─────────────┐
│  Next.js    │─────▶│  FastAPI     │─────▶│   Ollama    │
│  Frontend   │◀─────│  Backend     │◀─────│   (Local)   │
│ (Port 3001) │      │ (Port 8000)  │      │ (Port 11434)│
└─────────────┘      └──────────────┘      └─────────────┘
                            │
                            ▼
                     ┌──────────────┐
                     │   SQLite     │
                     │   Database   │
                     └──────────────┘

LLM Inference Flow

User types message → Frontend sends to /katja/chat
Backend receives request → Loads conversation history from SQLite
Prompt construction → Builds system prompt with grammar patterns + conversation context
Ollama inference → Sends prompt to local Ollama instance (http://localhost:11434)
JSON parsing → Extracts structured response (reply + grammar correction)
Database storage → Saves conversation + corrections to SQLite
Response → Returns to frontend with grammar feedback

JSON Format enforced via Ollama's format: json parameter:

{
  "reply": "Conversational response in target language",
  "correction": {
    "corrected": "Corrected version of user's text",
    "explanation": "Grammar explanation",
    "pattern": "Grammar pattern ID (if applicable)",
    "severity": "minor|major|critical"
  }
}

Technology Stack

Frontend:

Next.js (React framework)
TypeScript
Tailwind CSS
FontAwesome icons
react-markdown + KaTeX (LaTeX rendering) ⭐ NEW
remark-math + rehype-katex (Math typesetting) ⭐ NEW

Backend:

FastAPI (Python web framework)
SQLite with WAL mode (Write-Ahead Logging)
Pydantic v2 (validation)
SlowAPI (rate limiting)

LLM:

Ollama (local inference) - Primary
OpenAI GPT-4o-mini (cloud fallback) - Optional
Supports 14+ models including:
- llama3:8b-instruct-q4_K_M (general, 4.9GB)
- adrienbrault/saul-instruct-v1:Q4_K_M (legal, 4.1GB) ⭐
- meditron-7b (maritime medical, 4.1GB) ⭐
- biomistral-7b (biomedical, 4.1GB) ⭐
- glm4, qwen2:7b, gemma3:12b (multi-lingual)
Response length: 4096 tokens (configurable)
Temperature: 0.0 (deterministic for medical/legal accuracy)

Repository Structure

katja/
├── backend/
│   ├── core/
│   │   ├── database.py          # SQLite schema & connection
│   │   ├── db_operations.py     # CRUD operations
│   │   ├── llm_router.py        # Ollama + OpenAI integration
│   │   └── spaced_repetition.py # SM-2 algorithm
│   ├── flows/
│   │   └── teacher_flow.py      # Main conversation logic
│   ├── routers/
│   │   ├── teacher.py           # Chat endpoint
│   │   ├── reviews.py           # Spaced repetition API
│   │   ├── analytics.py         # Progress tracking
│   │   └── system.py            # Model switching, logs
│   ├── data/                    # SQLite database (auto-created)
│   ├── .env.example             # Configuration template
│   └── requirements.txt
├── frontend/
│   ├── components/
│   │   ├── Header.tsx           # Navigation, model selector, action buttons
│   │   ├── ChatWindow.tsx       # Grammar learning conversation UI
│   │   ├── GenericChat.tsx      # Generic/legal/medical chat with LaTeX rendering
│   │   ├── GrammarPanel.tsx     # Grammar corrections with localized labels
│   │   └── InputBox.tsx         # User input with multilingual placeholders
│   ├── config/
│   │   └── systemPrompts.ts     # ⭐ Centralized system prompts for all modes
│   ├── pages/
│   │   ├── index.tsx            # 🎓 Language learning page (default)
│   │   ├── chat.tsx             # 💬 General chat page
│   │   ├── legal-chat.tsx       # ⚖️ Legal chat page (Saul-7B)
│   │   ├── medical-chat.tsx     # 🏥 Maritime medical chat page (Meditron/BioMistral)
│   │   ├── console.tsx          # 💻 Console log viewer
│   │   ├── settings.tsx         # ⚙️ System prompt editor ⭐ NEW
│   │   ├── reviews.tsx          # Spaced repetition UI
│   │   └── stats.tsx            # Analytics dashboard
│   ├── utils/
│   │   ├── api.ts               # API client functions
│   │   ├── languages.ts         # Language metadata and labels
│   │   └── promptLoader.ts      # ⭐ Load system prompts (custom or default)
│   └── public/
│       └── screenshots/         # UI screenshots for documentation
├── launch-katja.sh              # Launcher script
├── CHANGELOG.md                 # Version history
└── README.md

Screenshots

1. Language Learning Mode

Multi-language grammar correction with real-time feedback in 7 European languages.

2. General Chat Mode

General-purpose AI conversations with model selection and structured responses.

3. Medical Chat Mode

Specialized medical queries using Meditron and BioMistral models.

4. Legal Chat Mode

English law queries using Saul-7B specialized legal model.

5. Console Mode

Terminal-style log viewer for monitoring all LLM interactions across modules.

API Endpoints

Language Learning

POST /katja/chat - Send message, receive grammar-corrected response
DELETE /katja/clear - Clear conversation history for language learning

General/Legal/Medical Chat

POST /katja/generic-chat - Generic chat with any model (supports LaTeX)
DELETE /katja/generic-chat/clear - Clear generic chat history

Console & Monitoring

GET /katja/console/history?type=language - Get language learning conversation history
GET /katja/console/history?type=general - Get general/legal/medical chat history
GET /katja/system/logs - Stream backend logs (real-time)

Model Management

GET /katja/models - List all available Ollama models
GET /katja/system/models - List available models with metadata
POST /katja/system/models/select - Switch active model

Spaced Repetition

GET /katja/reviews/due - Get corrections due for review
POST /katja/reviews/{correction_id}/review - Submit review result

Analytics

GET /katja/analytics/progress - Learning progress stats

Development

Running Tests

cd backend
pytest

Database Schema

SQLite database (backend/data/katja.db) auto-initializes with the following tables:

Language Learning Tables:

user_profile - User sessions, language preferences, and statistics
conversation_history - Language learning conversations with original and corrected text
grammar_corrections - Corrections with spaced repetition metadata (SM-2 algorithm)
grammar_patterns - Common grammar patterns for 7 languages (seed data)

Multi-Mode Tables:

generic_conversations - Shared conversation history for General Chat, Legal Chat, and Medical Chat modes
- Tracks session_id, role, content, timestamp, and model_used
- Enables context memory across conversation turns
- Queried by Console mode for monitoring

Features:

WAL mode enabled - Write-Ahead Logging for better concurrency (multiple reads during writes)
Session isolation - Each browser tab gets unique session_id stored in localStorage
Persistence - Conversations survive page refreshes and browser restarts
Auto-initialization - Tables created automatically on first startup

Database location: /backend/data/katja.db (auto-created, not tracked in git)

Troubleshooting

Backend won't start:

Check Ollama is running: ollama list
Verify Python dependencies: pip install -r requirements.txt
Check .env file exists (copy from .env.example)

No grammar corrections:

Verify model supports JSON output (use llama3:8b-instruct-q4_K_M)
Check backend logs in UI or /tmp/katja-backend.log

Frontend errors:

Ensure backend is running on port 8000
Check CORS settings in .env (add your frontend port)

License

Apache License 2.0

You are free to use, modify, and redistribute this code. See LICENSE for full terms.

Citation

S.M.Laignel. (2025). Katja: A Multi-Mode AI Chat Platform with Specialized LLMs for Language Learning, Legal Reasoning, and Medical Queries.
GitHub Repository. https://github.com/SL-MAR/katja

Name		Name	Last commit message	Last commit date
Latest commit History 20 Commits
backend		backend
frontend		frontend
screenshots		screenshots
.gitignore		.gitignore
CHANGELOG.md		CHANGELOG.md
Dockerfile		Dockerfile
FALLBACK_GUIDE.md		FALLBACK_GUIDE.md
LICENSE		LICENSE
README.md		README.md
docker-compose.yml		docker-compose.yml
docker-entrypoint.sh		docker-entrypoint.sh
launch-katja.sh		launch-katja.sh

License

SL-Mar/Katja

Folders and files

Latest commit

History

Repository files navigation

Katja — Multi-Mode AI Chat Platform with Specialized LLMs

Features

🎓 Language Learning Mode

💬 General Chat Mode

⚖️ Legal Chat Mode

🏥 Maritime Medical Chat Mode

💻 Console Mode

Core Features

⚠️ Important Disclaimers

Legal Chat

Maritime Medical Chat

Philosophy

Quick Start Guide

1. 🎓 Language Learning Mode

2. 💬 General Chat Mode

3. ⚖️ Legal Chat Mode

4. 🏥 Maritime Medical Chat Mode

5. 💻 Console Mode

Supported Languages

Prerequisites

Getting Started

Quick Start (Linux)

Manual Setup

Configuration

Backend Configuration

System Prompt Customization ⭐ NEW

How It Works

Architecture

LLM Inference Flow

Technology Stack

Repository Structure

Screenshots

1. Language Learning Mode

2. General Chat Mode

3. Medical Chat Mode

4. Legal Chat Mode

5. Console Mode

API Endpoints

Language Learning

General/Legal/Medical Chat

Console & Monitoring

Model Management

Spaced Repetition

Analytics

Development

Running Tests

Database Schema

Troubleshooting

License

Citation

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Contributors 2

Uh oh!

Languages

Packages