Link to Github

Objectives

By the end of the session you will have:

Used Kiro (CLI or IDE) to generate, scaffold, and debug a vision project on your own machine
Chained OpenCV operations on a live webcam feed (resize, threshold, mask, draw)
Run YOLO26n for real-time 80-class object detection, entirely on CPU
Seen SAM 3.1 segment objects from a text prompt (facilitator demo; optional to try yourself if bandwidth allows)
Built and presented a working webcam application of your own design

What we'll build

We cover the foundations together (OpenCV, real-time detection, segmentation), then you fork this repo, make a folder in projects/, and build whatever webcam app you want. Kiro writes most of the code; you steer it.

The seven demo apps in demos/ are there to run, read, and borrow from. You can rebuild one, mash two together, or ignore them and do something completely different. The prompt library (docs/prompts/) has the exact prompts that produced each demo, cleaned up so you can paste them into Kiro and swap in your own subject.

Technology	Role in this workshop
Python 3.10-3.12 and OpenCV	Webcam capture, frame manipulation, all on-screen drawing
Ultralytics YOLO26 (PyTorch, CPU)	Real-time COCO detection. The nano model (`yolo26n.pt`, ~5 MB) runs end-to-end without a separate NMS step and downloads itself on first use
Meta SAM 3.1	Text-prompted segmentation. Type "laptop" and get the pixel mask back. The model is ~6.5 GB, so we show it on a pre-loaded machine rather than downloading in the room
MediaPipe	Hand, face, and pose landmark tracking (21 / 478 / 33 points). Powers the gesture demos
Flask + flask-sock	Lightweight web framework already included. Every demo has a browser version you can run locally as a dashboard, and you can do the same for your own app. See prompt 07 in the library.
Kiro (CLI or IDE)	Local coding agent. Writes boilerplate, manages dependencies, helps you iterate
GitHub	You fork the repo, build in your folder, push when done

Reminders

Webcam preferred, not required. A webcam makes the demos more fun, but you can also work with uploaded images or video files. If you have a webcam, confirm it works before the session.
Local setup. We are using Kiro locally, CLI or IDE, your choice. Install whichever you prefer ahead of time so we don't lose session time to setup.
Public repo required. Your GitHub project must be set to public so it can be submitted.
Python 3.10-3.12 only. mediapipe has no wheels for 3.13+. Check with python --version before arriving.