Inside LLMs
How ChatGPT-like Models Work
- Saturday, 21 November 2026
- 9:30 am – 4:30 pm IST
- Main Seminar Hall, Block A, College of Engineering Kidangoor
What actually happens between your prompt and the reply? We follow a large language model from raw web text to a helpful assistant: tokenisers, pre-training, instruction tuning, preference optimisation, and the tricks that make inference fast. In the lab you'll train a tiny GPT on Malayalam poetry.
You'll be able to…
- Describe the LLM training pipeline end to end
- Explain context windows, temperature and sampling
- Train a character-level GPT from scratch
- Reason about cost, latency and model size trade-offs
Prerequisites
- Step 06 or transformer basics
What to bring
- Laptop (8 GB RAM recommended) and charger
- Python 3.11+ installed
- A Google account for Colab
- Curiosity — plenty of it
- Colab with GPU runtime enabled
Agenda
- 09:30Check-in & coffeeGrab your badge, find a seat, connect to Wi-Fi.
- 10:00The life of a large language model
- 11:30Short break
- 11:45Guided walkthroughLive coding with the speaker — follow along on your laptop.
- 13:00Lunch
- 14:00nanoGPT on Malayalam poetryWork in pairs. Mentors float between tables.
- 16:00Show & tellThree teams demo what they built.
- 16:30Wrap-up & next step preview
Speakers & mentors
Dr. Joseph Mathew
Principal Scientist, AI Safety, Indus Research Institute
Joseph leads evaluation research on large language models and advises policy bodies on responsible AI deployment.
Vishnu Prasad
Deep Learning Engineer, Monsoon AI, Kochi
Vishnu trains large models for a living and debugs exploding gradients for fun. Kaggle Master, PyTorch contributor.
Venue
Main Seminar Hall, Block A, College of Engineering Kidangoor
Kidangoor South P.O., Kottayam, Kerala 686583, India