May 2026 Meetup Minutes

May 23, 2026

Summary

The May 2026 ChennaiPy meetup began with an introduction to the history of Ethernet, followed by technical sessions on Machine Learning, Neural Networks, and Agentic AI. The event concluded with lightning talks on Linux graphics and Wayland, along with networking and a group photo.

Talk 1: Ethernet – A Brief History

Speaker: Shrayas

Shrayas opened the meetup by introducing the history of Ethernet and its significance in computer networking. He also spoke about Bob Metcalfe, the co-inventor of Ethernet, and highlighted his contributions to modern networking.

The session was followed by participant introductions.

Talk 2: Combat View – AI-powered Online Combat Sports Training

Speaker: Revanth

Revanth presented his Machine Learning project Combat View, an AI-powered platform for online combat sports training.

The project addresses several real-world challenges:

• Bridging the gap between physical gyms and remote training.
• Integrating real-time pose estimation into low-latency video calls.
• Preventing injuries caused by improper training forms.

He explained the project’s implementation using LiveKit for real-time communication and Ultralytics for pose estimation. He also introduced the Segment Anything Model (SAM) algorithm and concluded the session with a live demonstration of the project.

Talk 3: Understanding Neural Networks from Scratch

Speaker: Arunachaleswaran

Arunachaleswaran delivered an introductory session on Neural Networks and Machine Learning fundamentals.

Topics covered included:

• Neural network architecture.
• Hidden (middle) layers.
• Linear equations (y = mx + c).
• The concepts of weights and bias.
• Linear Regression using:

Output = Weight × Input + Bias

He demonstrated how prediction errors are minimized using Gradient Descent, discussed the role of Hyperparameters, and explained the Loss Function:

Loss = (Output − Prediction)²

Additional concepts included:

• Gradient Descent iterations.
• Single neuron architecture.
• Activation functions.
• Neural network configuration.
• The King–Queen vector analogy in embeddings.

The session concluded with an interactive Q&A.

Networking Session

Participants interacted with one another during the networking session, followed by a tea break.

Talk 4: PI (pi.dev) – An Open Source Agentic AI Tool

Speaker: Govindaraj

Govindaraj introduced PI (pi.dev), an open-source Agentic AI tool.

The session covered:

• Installation using CLI commands.
• Minimalist interface and workflow.
• Model selection.
• Login and subscription features.
• Solving the classic FizzBuzz problem.

He also discussed several key concepts:

• More tools require more tokens.
• PI focuses on simplicity compared to feature-rich AI assistants.
• Skills can be added as custom extensions.
• Skills support custom instructions.
• PI currently does not provide Plan Mode or Build Mode.
• Developers can build their own extensions because PI is open source.
• Self-modifying software.
• Customizing PI according to personal workflows.
• Introduction to the Mixture of Experts (MoE) architecture, including the 35B → 6B expert activation concept.

Lightning Talks

Linux Graphics Stack

Speaker: Praveen

Praveen continued his discussion on the Linux graphics stack.

Topics included:

• X11 and Wayland display servers.
• Framebuffer graphics.
• QEMU simulator.
• Rendering images in the Linux graphics pipeline.

X11 to Wayland

Speaker: Madhava Kumar

Madhava Kumar discussed the transition from X11 to Wayland.

Key points included:

• Scaling challenges in X11.
• X11 as the traditional display protocol.
• Wayland as the next-generation display protocol.
• Advantages of adopting Wayland for future Linux desktop environments.

Closing

The meetup concluded with a group photograph and ended at 6:10 PM.

Meeting minutes contributed by Micheal Anto Jerrish C (GitHub)