AI Engineering Meetup: Serve It Fast. Ship It For Real.
Details
Heads up: new start time. We now start at 4:30 PM, a half hour later than before, so we fit better with Industrious's schedule. If you show up at 4:00 out of habit, you'll be early.
This month, Sona Maniyan takes us inside vLLM and the memory trick that makes serving LLMs fast. Then Joshuha Owen shares what it actually took to put Rev.io's AI agent in front of paying customers.
Sona Maniyan: vLLM: LLM Serving with PagedAttention
Serving large language models at production scale is a different problem than training them. A big reason is the KV cache, which grows with every request and leads to fragmentation and over-allocation in traditional inference engines. PagedAttention borrows virtual memory paging from operating systems and stores KV cache blocks noncontiguously. Sona walks through how that works, then digs into vLLM, the open-source inference and serving engine built around it: a practical walkthrough, performance benchmarks (up to 24x higher throughput in the right comparisons), who's using it, and how you can contribute.
Joshuha Owen: From Chatbot to Product: Hard Lessons Shipping a Production AI Agent
Anyone can build an agent demo in a weekend. Almost nobody talks about what happens after, when it's in front of paying customers with real permissions, real data, and 300+ tools. This talk covers four hard-won lessons from shipping Revii, the AI agent inside Rev.io's Service and Revenue Operations SaaS platform:
- Identity is the foundation: where the agent inherits the user's actual permissions.
- Stay nimble on your harness: why we migrated from the OpenAI SDK to Mastra mid-flight.
- Tools at scale: you can't stuff 300+ tools in a context window.
- Generative UI: why that turns a chatbot into a product.
Agenda
Wednesday, November 11
🥨 4:30 PM · Doors open + snacks and networking
👋 5:00 PM · Welcome and intro remarks
🎤 5:10 PM · Sona Maniyan presents: vLLM: LLM Serving with PagedAttention
📺 5:40 PM · Joshuha Owen presents: From Chatbot to Product
💬 6:10 PM · Networking and discussion
🚗 6:30 PM · Go home and build
Address
Industrious, T3 Building
323 N Washington Ave, Suite 200 (2nd floor)
Minneapolis, MN 55401
Parking
Street parking is available, including metered spots in front of the building (2-hour limit). You can also park in Ramp C. From Ramp C, exit on the basement level on the northeast side (3rd Ave exit), walk around to the front of the T3 building, and come in the main doors.
Food + Drink
Food provided at 4:30 PM courtesy of our sponsor, Blank Metal. Drinks and the space courtesy of Industrious.
What to Bring
Just yourself. We've got everything covered.
Important to Know
Registration is free but required. Please RSVP so we can plan for capacity.
Accessibility
The T3 building is wheelchair accessible, with ramped curb access to the main entrance on N Washington Ave and an elevator to the 2nd floor.