🤖 Domain-Chatbot

A Local Domain-Based AI Chatbot

Developer: Valtry

---

📌 Overview

Domain-Chatbot is a fully local, domain-aware AI chatbot that changes its behavior based on the selected domain:

• 💬 Chat → Friendly, casual conversations
• 💻 Coding → Programming help with syntax highlighting
• 📚 Exam → Exam-oriented answers with structured points and formulas

The project runs 100% offline, uses Ollama for local LLMs, and supports real-time streaming, KaTeX math rendering, and code highlighting — all without any API keys.

---

✨ Why Domain-Chatbot?

Most chatbots treat every question the same.

Domain-Chatbot doesn’t.

• 💬 Chat mode → friendly, human-like conversation
• 💻 Coding mode → clean code + syntax highlighting
• 📘 Exam mode → structured answers + KaTeX formulas

All of this runs locally — no API keys, no internet dependency.

---

🖼️ Screenshots

💬 Chat Domain

💻 Coding Domain

📘 Exam Domain (KaTeX formulas)

---

🧪 Example Prompts

Chat

Explain black holes like I'm 10

Coding

Write a Python program to check palindrome

Exam

Derive Ohm’s law and explain with formula

---

✨ Key Features

• 🔁 Real-time streaming responses
• 🧠 Domain-based AI behavior
• 🧮 Automatic KaTeX rendering for formulas
• 🖍️ Syntax highlighting for code
• 💡 Centered suggestion prompts (one-time per domain)
• 🔒 Fully offline & privacy-friendly
• ⚡ Lightweight frontend (no frameworks)

---

🧠 Domain Behavior

Domain	Behavior	Special Handling
Chat	Friendly & conversational	Natural responses
Coding	Developer-focused	Highlight.js for code
Exam	Exam-oriented	KaTeX for formulas

---

🏗️ Architecture

Frontend (HTML + JS) ↓ FastAPI Backend ↓ Ollama Models (Local)

---

🧮 Smart Formula Detection (KaTeX Logic)

The chatbot intelligently decides what should be rendered as math.

Rules:

• Text inside ** ** with math symbols (= + - * / ^) → KaTeX
• Text inside ** ** without math symbols → Bold text
• Normal text remains unchanged

Examples: R = V / I → Rendered using KaTeX Resistance → Bold text Explanation: → Bold text (not KaTeX)

This prevents normal headings or explanations from being mistakenly rendered as math.

---

🛠️ Tech Stack

Backend

• Python 3.9+
• FastAPI
• Uvicorn
• Requests
• Ollama (Local LLM runtime)

Frontend

• HTML
• CSS
• JavaScript
• KaTeX (Math rendering)
• Highlight.js (Code highlighting)

AI Models (via Ollama)

• qwen2.5:0.5b → Chat
• qwen2.5-coder:0.5b → Coding
• gemma:2b → Exam

---

📂 Project Structure

Domain-Chatbot/ ├── backend/ │ ├── main.py │ ├── router.py │ ├── frontend/ │ ├── index.html │ ├── README.md

---

⚙️ Installation & Setup

1️⃣ Install Ollama

Download from: https://ollama.com

Pull required models:

ollama pull qwen2.5:0.5b ollama pull qwen2.5-coder:0.5b ollama pull gemma:2b

Start Ollama:

ollama serve

Install dependencies:

pip install fastapi uvicorn requests

Run the backend:

uvicorn main:app --reload

Backend URL:

http://127.0.0.1:8000

Frontend Setup

Open frontend/index.html directly in your browser OR Serve it using any static server No build tools required.

---

How to Use

1.Select a domain (Chat / Coding / Exam) 2.Use the centered suggestion prompts (shown once per domain) 3.Type or modify a prompt 4.Click Send 5.Watch the response stream live

---

👨‍💻 Developer

Valtry Built with curiosity, logic, and a focus on clean AI UX.

---

⭐ Support

If you found this project useful:

• ⭐ Star the repo
• 🍴 Fork it
• 🧠 Experiment with new domains

Your support really helps!

---

Final Words

This project is more than a chatbot — it’s a mini AI platform that demonstrates: Local LLM usage Streaming responses Smart content parsing Clean UI behavior Feel free to fork, extend, and improve 🚀

---

🔮 Roadmap

• Voice input/output
• Mobile-friendly UI
• More domains (Medical, Legal, Interview)
• Model selector UI

---