Name: In-person meetup: AI Tinkerers Ottawa v2.10.0 (last of 2024)
Start: 2024-11-19T17:30:00.000-05:00
End: 2024-11-19T21:00:00.000-05:00
Location: Ottawa, Ontario

We help companies go AI-native and create the ecosystem for AI Champions to thrive.

AGIV BUILDERS events here: https://luma.com/agiv-builders

AGI Ventures Canada

The ideal environment for agentic codegen (Harry Brundage - CTO @ Gadget)

Harry is a co-founder at Gadget trying to make app development suck less. Before Gadget, Harry worked at Shopify for a long-ass time scaling the core platform and the data infrastructure.

The holy grail of LLM codegen is app-at-a-time generation -- whole application backends, frontends, integrations, features, and hosting summoned with a few short sentences. We can barely get a passing grade on PR-at-a-time generation benchmarks though, due to both model abilities and agent architecture -- the agents are still babies. At Gadget, we think we accidentally spent the last 4 years building the ideal environment to overcome both these issues, but... we aren't quite sure. The baby needs a cradle. Join Harry on an adventure in building a codegen agent that's fundamentally different than the one's you've seen before.

Unlocking the world's fastest inference (Maria Mihu Gleb Kazantaev, members of technical staff @ Cerebras Systems)

Only three months ago, we launched Cerebras Inference, the fastest AI inference solution in the world. But how can AI tinkerers leverage this powerful service, and why should they care? In this demo, Cerebras ML Engineers will guide you through the many ways you can harness the power of Cerebras inference in your own projects. We will go over our developer tools, third-party integrations, and community projects. When we are done presenting, you should all be ready to get started on building any fast inference applications that you can dream of.

Building complex machine vision video applications with Workflows (Sachin Agarwal - infra @ Roboflow)

Sachin Agarwal leads the infrastructure team at Roboflow that manages 1000s of machine vision trainings and millions of inferences every day.

Witness the power of Roboflow Workflows as we transform a simple object detection model into a highly accurate system using SAHI.

Explore the versatility of our workflow blocks, including zero-shot

detection with Microsoft Florence 2 and YOLOv8. Dive into  video

analysis and discover how Roboflow is revolutionizing computer vision development. Join us on this journey and unlock your potential to build cutting-edge vision applications with Roboflow.

Decompiling a model from it's weights (Alice Rigg - EleutherAI)

Even though ever larger and more capable machine learning models are being deployed in real-world settings, we still know concerningly little about how they implement their many impressive capabilities. One emerging approach for understanding the internals of neural networks is mechanistic interpretability (MI): reverse engineering the algorithms implemented by neural networks into human-understandable mechanisms. Though this is an ambitious goal, in the past two years, MI has seen rapid progress. In this talk I'll highlight lessons learned in these years, what applications has MI enabled that were previously not possible, and how long the road is to come. I'll close with a teaser of what my research group has been working on, relating to the title of the talk, and how it addresses key bottlenecks in the field.

LLMs in the User Agent (Mauve Signweaver)

Mauve is a tech consultant focused on bleeding edge software and decentralized systems. Their latest passion has been locally runnable LLMs and low power machine learning.

This talk will go over the LLM integration in the Agregore browser and how web apps could make use of browser-supplied AI APIs instead of relying on cloud or WASM.

We have a dedicated table for showing off and discuss with other builders during networking time.

Our group welcomes developers of all backgrounds. If you’re passionate about the technicalities behind developing AI systems, this is a home for you. Our promise is to provide a stage for open-source AI developers to share code and the lessons they’ve learned from experimentation.