NornNet

Visit NornNet. Where the Fates Weave Destiny with a Touch of AI Magic - A private AI ChatBot that rocks!

Team Members

• Eliezer Lamien
• Joe Scott
• Owen Osmera
• Shawn Noon

Flask Tutorials

• Chapter.16_Web_Development.pdf
• Flask Unit Converter
• Flask Blog Tutorial

AI Private ChatBot Research

• LM Studio Tutorial: Run Large Language Models (LLM) on Your Laptop
• Network Chuck Private AI
• How to Build a Local AI Agent With Python (Ollama, LangChain & RAG)
• The Ultimate Guide to Local AI and AI Agents (The Future is Here)
• Part 1 - Run Ollama with Python in 5 Minutes (Full Setup Guide
• Part 2 - Local Ollama Chatbot in Python
• Demo Creating an App Using Ollama OpenAI Python Client
• How to Create Your Own AI Chatbot Like DeepSeek with Ollama in Python Flask
• Learn Ollama in 15 Minutes - Run LLM Models Locally for FREE
• How To Prompt AI in 2026 (according to experts)-NetworkChuck

Personal Research

• Integrate an LLM into flask with python https://github.com/ollama/ollama-python
• Create, read, and write to a PDF file https://github.com/py-pdf/pypdf
• Chat history/Memory/Context https://github.com/digithree/ollama-rag

Create Pull Request on NornNet Github

1. Fork the NornNet Repository.
2. Pull local copy to your computer.
3. Make code changes.
4. Commit --> then Push to your repository
5. If your repository is not up to date --> Synchronize Changes.
6. Create a Pull Request.

AI ChatBot Project (Private AI options)

• Please watch: Video about Guild Project (10-16-25)

Python Flask project to create a web chatbot. Private mode: host an open-source model yourself (local server, private cloud, or on-prem) for full control over data and privacy.

Project Outline: Flask Chatbot (Private AI)

1. Project Goal 🎯

Build a reliable web chatbot. The app should let users exchange messages with an AI model and see streamed responses.

• Privacy-first solution running a private model server (containerized) with secure access and logging controls.

2. Key Technologies

• Backend: Python, Flask, Waitress (chat endpoint and API connector)
• Frontend: HTML, CSS, JavaScript
• Private AI: local model server (lightweight runtime like llama.cpp / GGML for CPU quantized models)

---

3. Architecture & Implementation Plan

High level: Client (browser) -> Flask app -> (Private model server) -> Flask -> browser.

The implementation is split into phases so the team can ship incrementally.

Phase 1: Setup Flask Server & Ollama Backend

✅ Install ollama on Windows 2025 Server

1. Download the latest Ollama Release (e.g., ollama-windows-amd64.zip)

2. Extract the zip file --> place all files in c:\ollama

3. Add to system environment variables: c:\ollama

4. Verify installation: Open a Command Prompt:

5. Verify that gemma3:4b is installed as the model

   ollama --version

6. Pull a Model to test with:

   ollama pull gemma3:4b

✅ Set Up Ollama as a Windows Service (using NSSM)

This ensures Ollama starts automatically with the server and restarts if it ever crashes, without requiring a user to be logged in.

Step 1: Download NSSM

1. Go to the NSSM download page.
2. Download the latest release (e.g., nssm-2.24.zip).
3. Extract the ZIP file to a permanent location on your server, like C:\nssm

Step 2: Configure the Ollama Service with NSSM

1. Open an Administrator Command Prompt or Administrator PowerShell.

   nssm install Ollama

This will open the NSSM service installer GUI.

Application Tab:

1. Path: Click the ... button and navigate to your ollama.exe file.

2. Startup directory: This should automatically populate to the correct folder.

3. Arguments: This is the most important part. Enter:

   serve

This tells Ollama to run as a server.

4. Log on Tab: For a server, it's best to run this as a Local System account. Select the

Local System account

radio button. This allows the service to run even when no user is logged in.

5. Environment Tab: This is critical for allowing your Flask server to access Ollama.
6. In the Environment variables box, add the following key-value pair:

OLLAMA_HOST=0.0.0.0

This tells Ollama to listen for requests on all network interfaces, not just localhost.

7. Click the Install service button. You should see a "Service 'Ollama' installed successfully!" message.

Step 3: Start the Service

1. Open the Windows Services app (run services.msc).

2. Find your new service named Ollama.

3. Right-click it and select Start.

4. Verify it's working: Open your server's web browser and navigate to:

   http://localhost:11434

5. You should see the message "Ollama is running."

The Ollama server is now running as a persistent service on port 11434, accessible from other machines on your network.

☐ Todo — Draft README 🔄 In progress — Add examples and docs ✅ Complete — Initial release

Flask Web Server

✅ Complete — Initial release

• Create main_app.py with routes for the homepage (/) and chat (/chat).
  • Dynamic base path support: set BASE_PATH=/nornnet in .env for production; leave unset locally to serve at root.
• Build index.html with a chat history, input, and send button.
• Implement minimal CSS to make the UI usable.
• Download the student handbook onto the server for the AI to reference
• Create virtual environment

  • Install python ollama library:

    pip install ollama

Phase 2: Frontend-Backend Communication

✅ Complete — Initial release

• Use main.js to send user messages to /chat and append both user and bot messages to the DOM.

Phase 3 — Private AI Integration (self-hosted)

✅ Complete — Initial release

• CPU-only machines: prefer quantized models (GGML / llama.cpp) or small transformer models.
• Create a model server and expose an internal HTTP endpoint.
• Secure the private server: mTLS / API key, and run behind an IIS server which runs waitress

Tasks for the week of 11/10/25 - 11/16/25

• Make the ai function a class for expandability.
  • DO NOT chage the AI function file only make a new file for the class.
  • SetUserQuestion
  • GetUserQuestion
  • GetAIResponse
  • SetAIPrompt
  • GetAIPrompt
• Find a way to get a pdf reader in python and integrate that into the code.
  • How do we add the student handbook as context for the ai?
  • This can be a function for class we just need a way to read the file to create words for prompting the ai.
• Find out if context can be enabled on ollama.
  • Integrate this into the code so the AI can have context
  • Research using CAG or RAG
• Add github repository address to docs page.
• Add responsive design to user interface. The interface should resize for different devices. Look at ChatGPT for an example
• • Add support for streaming responses if the model server supports it. This may provide a faster start to the response and a better user experience.

Phase 4 — Privacy, Logging & Data Handling

☐ Todo

• Decide what to log (timestamps, anonymized session id, message length). Avoid storing PII by default.
• Add a user-facing privacy notice and an opt-out for logging.
• Implement retention policy and a script to sweep/delete logs older than X days.

Phase 5 — Deploy & Ops

☐ Todo

• Add health checks, simple metrics (request counts, success/failure), and basic monitoring instructions.

---

4. Task Breakdown

This plan maps team roles to implement the privacy-oriented private mode.

Frontend (UI/UX)

• Team 1 (HTML & Structure): index.html — chat layout, message list, input form.
• Team 2 (CSS & Interactivity): style.css and main.js — chat bubbles, responsive layout, DOM updates, optimistic UI.

Shared goal: Deliver a clean, usable chat UI with graceful error states.

Backend (Flask Core)

• Team 3 (Server & Routing): app.py — create routes, session handling, simple persistence for sessions.
• Team 4 (Request & Response Handling): Implement /chat logic that validates input, forwards to the connector, and returns JSON or streaming responses.

Shared goal: Reliable, well-documented endpoints and simple persistence for conversation history.

AI Integration & Ops

• Team 5 (Private AI & Security): Implement local/private model server wiring and implement logging & retention rules. If hardware allows, tune for streaming and lower-latency inference.

Optional extras for the team:

• Add unit/integration tests for the connector code.
• Add a small evaluation page that replays test prompts and shows model outputs side-by-side.
• Add simple cost/latency telemetry so the team can compare different models.

---

License

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

Copyright (c) 2025 WNCC IT Program