Gamer's Local LLM Outperforms Calendar in Scheduling Tasks

Samarveer, a gamer and literature graduate, was dissatisfied with his calendar's ability to structure his day. He decided to use his local LLM to schedule his tasks more effectively.

He connected his LLM, Qwen3.5, to Google Calendar using the llama.cpp library and a Python script that took a task list and estimated duration from Notepad. The script then turned this information into a schedule for the day.

The initial attempt with Qwen3-8B failed to include some tasks, but after modifying the Python script to restrict certain events, GPT-OSS-20B was able to create an efficient schedule. Samarveer found that his LLM was more effective at scheduling his day than he could be himself.

The experiment showed that even with a well-designed Python script and a suitable local model, the Google Calendar API proved to be challenging to use.