Dev Log #6 – Local AI Integration with Ollama

Overview

Today I addeed a completely local AI model using Ollama. I researched local large language models and decided to integrate the Llama 3.2 3B model through Ollama.

What I Worked On

• Installed Ollama and downloaded the llama3.2:3b model.
• Fixed the issue where PowerShell could not recognize the ollama command by updating my system PATH.
• Learned that Ollama automatically starts its own server, so running ollama serve again caused a port conflict.
• Created a Node.js backend using Express and CORS.
• Connected the backend to Ollama’s local API.
• Updated the React frontend to send AI requests to my local backend instead of OpenRouter.
• Removed the dependency on paid AI APIs.

Challenges

One of the biggest challenges was getting the backend running correctly. I accidentally tried to run server.js before creating the file, which caused a MODULE_NOT_FOUND error. After creating the server and installing the required packages, I was able to start it successfully.

Another challenge was that the AI initially hallucinated events like holidays and birthdays that weren’t actually on my calendar. To fix this, I rewrote the prompt to clearly instruct the model to only use the calendar events provided and never invent additional information.

Improvements

I redesigned the AI briefing prompt so it now:

• Uses only events from the next 7 days.
• Creates a structured daily briefing.
• Highlights upcoming due dates, tests, meetings, and deadlines.
• Suggests what to focus on for the day.
• Ends with a short motivational message.
• Avoids generating fake events.

Result

The AI summary feature now runs entirely on my own computer with:

• No API keys
• No usage limits
• Faster response times
• Better privacy because calendar data never leaves my computer.

The morning briefing is much more useful and feels like a real student assistant instead of a generic chatbot.