We evaluated this pipeline on corporate SEC 10-Q financial filings. The structural context allowed smaller, cost-effective models like GPT-4o-mini to jump from 36 to 56% accuracy, all while adding only 0.05 seconds of latency.

This approach solves a major challenge in document extraction: lost context. By hyphenating entities with their source document titles before embedding them, the system preserves critical document-level hierarchies that usually disappear during standard text chunking.

To address this, we developed an Entity-Based Knowledge Graph RAG architecture (KG-RAG). Instead of just storing chunk embeddings, this system maps key entities and their relationships into a structured graph, anchoring every connection back to the source text.

Most retrieval-augmented generation (RAG) systems treat text like disconnected blocks. They excel at retrieving paragraphs that sound relevant, but struggle when an answer requires connecting facts scattered across multiple pages or fiscal quarters.

The outcome? Used as a reward signal with reinforcement learning, 3DSPA measurably improves physical realism in generative video models. Gravity works properly and objects stop passing through solid barriers.

To bridge this gap, Allen introduced 3DSPA: a 3D Semantic Point Autoencoder. It predicts 3D point trajectories to measure physical realism in generated video, detecting physical rule violations just like a human infant would.

Modern models excel at knowledge representation but struggle with physical reasoning. Humans solve novel physical problems easily via accurate mental simulation, a capacity current generative models lack.

How does cognitive science help us build better AI? At our 2026 AI Summit & Career Fair, Vector Faculty Member Kelsey Allen broke down how human cognition exposes the limits of current learning systems and how we can fix them. 🧵👇

That’s a wrap on the 2026 AI Summit and Career Fair! 🚀 From main stage insights to an engaging Career Fair, today showed why Ontario leads in AI innovation. Huge thanks to our speakers, sponsors, and all our attendees!

How do you break into AI? Our panel with talent leaders from Bell, Accenture, Layer 6, and Emterra broke down exactly what enterprise teams look for and how candidates can break through in a complex market.

Deploying AI in finance means high-stakes decisions at scale. Golnoosh Samei (RBC Borealis) took us under the hood of ATOM, RBC’s Large Transaction Model (LTM) that powers agentic workflows, moving enterprise AI far beyond the basic chatbot.

How do cognitive science and AI shape each other? Vector Faculty Member Kelsey Allen discussed mental simulation and how tools like 3DSPA detect physical rule violations, measurably improving the physical realism of generative video models.

The morning sessions at the AI Summit and Career Fair delivered deep insights into AI research and industry adoption. Here’s what we heard from the main stage! 👇🧵

We are incredibly grateful to the Government of Canada and the Ontario Government for their continued partnership in building and supporting our AI ecosystem. Now let’s spend the day defining the future of AI together.

Minister Solomon addressed the AI talent in the room: "The distance between ideas and execution has never been shorter. You have the most powerful tools in history to solve problems."