Seattle AI Hack Night Recap

Some folks from the Hypermode team along with our friends at Boundary, Neo4j, and Union AI joined about 200 members of the Seattle AI community for a Seattle AI Hack Night event. The night kicked off with a handful of lighting talks showing the latest developer tools from Hypermode, an overview of BAML, how to leverage GraphQL with Neo4j, and orchestrating AI pipelines with Union AI. Then the community got down to business by forming teams to build and demonstrate fun AI apps using these tools.

Fortunately we captured recordings from the event so we can share them with you here!

Introducing BAML: Making LLMs Reliable in Your Codebase | Trip Planner Demo

In this talk, Vaibhav Gupta, co-founder and CEO of Boundary, introduces BAML - a framework that makes working with LLMs more reliable and maintainable in your codebase.

Vaibhav demonstrates how BAML solves the same problems for LLM integration that React solved for web development - providing structure, type safety, and better developer experience. The presentation includes impressive demos of a Tesla trip planner that generates routes with charging stops, and an interactive recipe generator.

Learn how BAML enables you to:

• Write type-safe LLM functions with guaranteed output schemas
• Debug your prompts with real-time visualization
• Build interactive AI applications that feel like native experiences, not ChatGPT wrappers
• Test your LLM code with confidence

Check out the open-source BAML framework and see how it transforms AI development from string-based prompts to maintainable, production-ready code.

Introduction to Flyte: Orchestrating AI Pipelines with Union AI - Lightning Talk

In this lightning talk, Sage Elliott from Union AI introduces Flyte, an open-source orchestration tool for complex AI models and data pipelines.

Learn how Flyte enables efficient ML workflows with containerized tasks, caching, and versioning through simple Python decorators. Sage demonstrates how to build reproducible pipelines for training language models, showcases the Actors feature for long-running stateful containers, and explains how organizations like LinkedIn have adopted Flyte for their LLM training workflows. Perfect for AI engineers looking to streamline experimentation and deployment with minimal infrastructure overhead.

Do You Really Need All That Prompt? Optimizing AI with Neo4j | GraphQL Plain Text Demo

In this talk, Dan Starns demonstrates how to build more efficient AI tools by reducing prompt sizes when working with large language models. He showcases GraphQL Plain Text, a tool that converts plain language into GraphQL queries, and explains the challenges of feeding entire schemas to LLMs.

Learn how Neo4j's graph database can be leveraged as a vector store to create a more precise RAG (Retrieval-Augmented Generation) approach, reducing hallucinations and resource waste. Perfect for developers interested in AI optimization, GraphQL, and practical solutions for working with large schemas.

Dan also shares code examples and introduces GQL PT Modus for Hypermode users, along with a preview of a platform designed for AI-generated queries and team workspaces. A must-watch for anyone building AI tools who wants to improve efficiency and accuracy!

Building Model-Native Apps with Hypermode: A Developer's Guide to Serverless AI Tools

In this talk, Will Lyon from Hypermode presents a comprehensive overview of building model-native app using open-source tools. He introduces Modus (a serverless API framework), Dgraph (a real-time graph database), Badger (a key-value store), and Ristretto (a cache library) - all designed to simplify AI-powered application development.

Learn how Modus leverages WebAssembly to support polyglot development with AssemblyScript and Go, automatically generating GraphQL APIs from your functions. Will demonstrates how to get started with a simple project and showcases a more complex example integrating Neo4j for vector search with movie recommendations.

Perfect for developers looking to incorporate AI models, knowledge graphs, and serverless architecture into their applications. Check out Hypermode's Modus Recipes repository for starter kits and examples to jumpstart your next project!

Hack Night Presentations

• FlavorFindr - Using Modus, Hypermode hosted embeddings model, Neo4j, and Next.js to build a full stack restaurant recommendation app using vector search of restaurant reviews and descriptions.
• MomAI - The AI Reminder Bot That Acts Like Your Mother - An AI agent that integrated with your calendar to remind and motivate you of upcoming events, using AI prompts via BAML to customize the type of reminder.
• Reddist AI - Cure the Reddit data firehose by using an AI agent to process and analyze Reddit content by generating customized newsletters delivered via email.
• Developer Analysis - Analyze GitHub repositories using AI to determine the health of the repo, if it follows best practices, and generate practical suggestions for improvements.

And the winner was.... MomAI!

At Hypermode we love connecting with our community through in person events. If you haven't already, be sure to subscribe to the Hypermode Luma event calendar to be notified of upcoming events in your area and online! Hope to see you at one of our events soon!