Last week, we dove into Deep Research through the API and built Circle of Mages 🧙‍♂️: A multi-agent AI research system that leverages both PhD-level and Masters-level AI agents to conduct comprehensive industry surveys and synthesize findings into academic survey papers.
Prompt engineering was only the warm-up—Context Engineering is now the main event. In this session, we’ll show why expertly curating an LLM’s context window is the essential skill for modern AI builders. From Retrieval-Augmented Generation and agent design to in-context memory and tool orchestration, you’ll see how today’s best practices are shifting—and why leaders like Andrej Karpathy and Tobi Lütke are all-in.
We’ll break down four core tactics—write, select, compress, and isolate—then bring them to life with a live demo of an agentic RAG application. If you’re shipping production-grade LLM systems in 2025, mastering Context Engineering is your next-level unlock. Join us and level up.
Guardrails aren’t optional anymore—they’re the foundation of any safe, dependable LLM product. In this session, we’ll demystify what guardrails are, why they’re mission-critical, and how to weave them through the entire AI pipeline—from pre-generation checks to post-deployment monitoring. We’ll walk through the AI Guardrails Index, spotlighting 20-plus solutions that tackle jailbreak defense, PII redaction, hallucination control, and more. Finally, we’ll build a production-ready, guardrail-hardened app live on stage. If trusted AI is on your 2025 roadmap, this is the blueprint you’ve been waiting for.
In this session we explore two of the most recent developments in the field: Agent-to-Agent protocol (A2A) and Model Context Protocol (MCP). We dive in to explore how to unpack A2A’s open-standard “operating system” for agent teamwork, show how it complements the Model Connector Protocol (MCP) “USB-C for LLMs,” and walk through real-world examples like a multi-agent travel planner.
COMMUNITY HIGHLIGHTS
🚀 Small Language Models are the Future of Agentic AI!
Adrian V. a.k.a "Fluff" gives you a hands-on demonstration of the Nvidia x Georgia Tech paper "Small Language Models are the Future of Agentic AI" paper. ​ The paper advocates for a paradigm shift from "LLM-centric" to "SLM-first" architectures, where intelligent routing systems automatically select the right model for each task—combining the precision and efficiency of SLMs with the generality of LLMs only when needed. Check out his app to test it out!
Cohort 6 Highlight: 🚬 Planolyzer by Michelle Boyer — AI Shelf Scans That Stop Out-of-Stock Losses
Have you ever made a run to the gas station for your favourite brand of cigarettes only to find they're out? Meet Planolyzer. Planolyzer is a multimodal AI tool that lets truck-stop retailers snap shelf photos, instantly detects empty slots, misplaced SKUs, and planogram violations, and sends actionable restock prompts.