Speakers
Description
You can build, customize, and ship a working AI agent with no API key, no server, and not a rupee spent. In this 90-minute hands-on workshop, you'll do exactly that.
We'll use WebLLM (an open-source browser runtime from MLC-AI), Google's open-weight Gemma 2 2B model (quantized to 4-bit, roughly 1.5 GB), and WebGPU, the browser's native GPU compute API, to run real LLM inference entirely inside a browser tab. The whole stack is open source, and the only tools you need are an Ubuntu laptop and a WebGPU-capable browser.
The workshop moves through four phases.
Phase 1 covers the AI loop: load a quantized language model from a CDN import, make your first chat completion, add streaming, and wire up multi-turn conversation memory, all in a single HTML file with zero build tooling.
In Phase 2, you make it real. Turn the raw text loop into a styled chat application with message bubbles, auto-scroll, a loading progress bar, and input state management.
Phase 3 is where you make it yours. Customize your agent's persona and system prompt (fitness coach, travel guide, language tutor, your pick), then build a research agent that calls real tools (Wikipedia, weather APIs) through an agentic loop. The model decides when it needs more information, calls a tool, reasons over the result, and loops until it has a grounded answer.
Phase 4 ships it: deploy your agent to GitHub Pages in three steps. You leave with a live URL.
Every attendee walks out with a working, deployed, zero-cost AI agent, plus the mental model to extend it with new tools, bigger models, and richer UIs.
Prerequisites: a laptop with GPU enabled, checked via chrome://gpu), a text editor, and basic familiarity with HTML and JavaScript. No ML or AI experience required . Please check the pre-workshop in presentation material
Summary
Build and deploy a working AI agent that runs entirely in your browser, with no API keys, no servers, and no cost. This hands-on workshop uses WebLLM, Google's open-weight Gemma 2 model, and WebGPU to take you from a blank HTML file to a live, tool-calling AI agent on GitHub Pages in 90 minutes.
Speaker Bio
Nishant Kumar Thakur is a Senior Computer Scientist at Adobe, where he works on AI platform engineering, browser-native ML inference, and agentic AI pipelines. His work spans WebGPU, ONNX Runtime Web, WebLLM, and MCP-based tooling. He holds an MS in Data Science from Deakin University and a patent. He is an active speaker, workshop facilitator, and hackathon judge across communities including AI House, GDG, IDC and MLX India.
| Speaker LinkedIn / Socials | https://www.linkedin.com/in/nishantkthakur/ |
|---|---|
| Difficulty | Beginner |