Chicago AI, ML, and Computer Vision Meetup - July 23, 2026

Join our in-person meetup in Chicago to hear talks from experts on cutting-edge topics across AI, ML, and computer vision.

Data curation is often described as drudgery, but an agent can turn it into an interactive, puzzle-solving process. This talk walks through a data-centric computer vision workflow built on the FiftyOne platform and the FiftyOne Agent, which uses skills (step-by-step instructions for tasks) and the Model Context Protocol (MCP) to handle dataset management, model inference, and brain computations through natural language — querying metadata, visualizing class distributions, and running complex operations without writing manual code.

The demo applies the workflow to a crop and weed dataset of 8,034 images and more than 112,000 labels (bounding boxes, keypoints, and segmentation masks). The agent diagnoses severe class imbalance and duplicate frames, builds interactive histograms and dashboards to surface domain shifts like lighting variation, finds invalid annotations so they can be corrected directly in the FiftyOne App, then automates exporting the curated dataset, fine-tuning a YOLO model, and evaluating its performance. The result is a repeatable pattern — ingest, explore, improve, fine-tune, evaluate, iterate — with a human in the loop through an intuitive, agentic interface.

Radiologists face mounting report volumes and rising burnout, creating demand for assistive tools that can draft structured findings directly from imaging. We fine-tuned Qwen2.5-VL with LoRA on roughly 5000 paired lumbar spine MRI studies and reports, training the model to generate spinal level-specific findings evaluated with lexical, semantic, and clinical metrics.

We are extending the work with self-supervised pretraining on an additional 20000 unlabeled studies to build domain-specific backbones for downstream tasks including lumbar spine MRI segmentation and classification. The talk shares current results, challenges faced, and why evaluating structured radiology reports is harder than standard metrics suggest.

Stereo depth estimation is a core perception capability in robotics and autonomous systems, converting rectified stereo image pairs into dense disparity and depth maps. While modern deep stereo methods achieve strong accuracy, state-of-the-art models, especially transformer-based architectures often incur high computational and energy costs, limiting deployment on resource-constrained devices.

This studies stereo depth estimation and proposes an efficiency-oriented modification to a transformer-based stereo pipeline by incorporating Walsh–Hadamard Transform (WHT) operations into the feature extraction stage. Specifically, we experiment with a WHT-based convolutional substitute (WHTConv2D) to reduce multiply-accumulate operations while preserving representational capacity via structured ±1 transforms.

We inferenced classical and neural stereo models on a specific dataset to compare, culminating in an STereo TRansformer (STTR) baseline and a WHTConv2D-enhanced variant. The proposed design achieves an observed 18.33% efficiency improvement relative to the baseline configuration while maintaining competitive long-range disparity accuracy.

Developing Agents Skills can help document domain expertise, create repeatable recipes, and even increase adoption of your tooling. But, understanding how well agents can actually use these skills to generate appropriate code is not easy. Scale that up across agents and runtimes, and the problem space explodes!

Cultivar is a packaged developed at Pinecone aimed at solving these problems, providing a CLI over sandboxed runtimes for agents, and a framework for specifying tasks and graders. This talk will cover what skills are, what Cultivar is, how it works, and what we learned at Pinecone using it to develop skills.

The talk will focus on the importance of clearly defining a specific problem and a use case, how to quantify the potential benefits of an AI solution in terms of measurable outcomes, evaluating technical feasibility in terms of technical challenges and limitations of implementing an AI solution, and envisioning the future of enterprise AI.