Final-year project · PUCIT

Activity recognition with LLMs

My final-year project at PUCIT: a web app that recognizes human activities from smartwatch, phone and glasses data, using pre-trained LLMs fine-tuned for the task.

Project
Final-year project, BS IT
University
PUCIT, University of the Punjab
Finished
2025
Stack
.NET, Python
Full title
Recognition and Demonstrations of Human Activities Leveraging LLMs
  1. inputWatch, phone, glasses
  2. encodeInput embeddings
  3. modelFine-tuned LLM
  4. outputRecognized activity
human activity recognition from multi-device wearable data
From three devices to one recognized activity.
01

The project

The app takes in data from three devices at once, a smartwatch, a phone and smart glasses, and recognizes the activity being done.

It was my final-year project for the BS in Information Technology at PUCIT, University of the Punjab.

02

How it works

I integrated pre-trained LLMs through input embeddings, so sensor data reaches the model in a form it can work with, then fine-tuned them for activity recognition. The model sits behind a web app built with .NET and Python.

  • .NET
  • Python
  • Pre-trained LLMs
  • Input embeddings
  • Fine-tuning
Next case studyCoursera CloneA learning platform with dashboards, subscriptions, quizzes and certificates.