Chicago AI, ML, and Computer Vision Meetup

Hear first-hand how data, analytics, and AI are innovating every business industry. Ari will lead a lively discussion on the latest in AI: data intelligence platform on a data lakehouse architecture, the importance of collaboration among humans and their data, and a demo on creating a fully governed LLM chatbot in under a minute!

Agent architectures are evolving toward a reinforcement learning mindset. The “observe-plan-act” ReAct paradigm, for instance, matches the RL paradigm almost exactly. This convergence naturally raises the question: could RL techniques actually improve our agentic systems? In this talk, I’ll explore the answer to that question. I’ll show how to leverage external signals as an implicit reward function, thereby allowing agents to autonomously learn from their environment. I’ll also discuss how inverse reinforcement learning (IRL) techniques can infer a reward function implicitly from observing human expert demonstrations. Using a human-in-the-loop tech-recruiting platform as our guide, we’ll see how these techniques can fine-tune agent behavior through both external cues and expert feedback.

Building multimodal RAG applications can be tricky, especially for video search applications. When working with recorded presentations, information can exist on screen and in spoken audio, which requires models and processing techniques that take advantage of this asymmetry.

In this talk, Arjun Patel (Developer Advocate at Pinecone) will demo an application built with Pinecone’s vector database and Claude models that allows for retrieval augmented generation over webinar videos.

The key trick is applying Anthropic’s technique of contextual retrieval to preprocess and enrich the video data into text for semantic search, along with a multimodal RAG step with Claude. The talk will cover how and why contextual retrieval works, preprocessing video data into image text pairs, and leveraging the multimodal nature of Claude to further refine response quality.

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.