Reliability Calls #2: Building Business-Aware AI — From Clean Semantics to Continuous Context

Every month, we unpack what it really takes to build Reliable AI — AI that doesn’t just demo well, but works reliably in production, handles edge cases, and earns trust over time.

​Building Business-Aware AI — From Clean Semantics to Continuous Context

In this edition, Jilong Kan (Sr. Director of Engineering, Flexport) joins us for fireside chat of Flexport’s newly launched Insights Builder — an AI-powered tool built on top of a clean, unified semantic foundation that lets users ask plain-English questions and get real-time supply chain insights and build custom dashboards on-the-fly. But for most teams, that kind of data foundation is years away.

So what do you do in the meantime? Even Flexport’s team is evolving — adding new context, new data sources, and constantly adapting as business definitions shift.

After the fireside chat, the PromptQL team will demo how CompanyQL, our agentic semantic layer, helps teams codify their business logic and context into something LLMs can understand and execute with reliability — even in messy, imperfect environments.

You’ll leave with practical ideas for shipping reliable AI today, even if your data isn’t “AI-ready.” ​Who should attend:

AI platform leads, semantic layer owners, and technical decision-makers working on GenAI systems that need to operate with real-world reliability and context.

This session will feature:

​​Shipping Business-Aware AI: How Flexport Built Insights Builder - Fireside chat with Jilong Kan (Sr. Director of Engineering, Flexport)

Flexport’s Insights Builder gives customers the ability to transform complex data interactions, making insights and generating custom dashboards as accessible as asking a natural language question. But under the hood, this system is powered by a clean semantic foundation, meticulously built to support accuracy, extensibility, and customer context.

• How the team handles evolving business definitions and continuous context

• ​The tooling they’ve built for monitoring accuracy, both online and offline

• ​Lessons learned from turning a great AI idea into a trusted product

• ​It’s not just about being AI-ready — it’s about staying AI-reliable as the ground shifts beneath you.

​The Other 99%: Building Reliable AI Without Waiting for Clean Data

Let’s be real — clean semantic layers are a dream come true. And even when you get there, the context is constantly shifting: new data sources, evolving definitions, and changing business logic.

So how do you build AI systems that stay reliable through all of that?

​In this quick lightning talk, we’ll show how PromptQL helps teams ship CompanyQL — a domain-specific planning language that encodes your unique semantics, tribal knowledge, and business logic into something LLMs can execute deterministically.

You’ll see:

• ​A live demo of our Agentic Semantic Layer building meaning on-the-fly

• ​How Reliability Scores help you track and trust AI performance

• Why this works — even if your data’s messy and your definitions are still evolving

Whether you’re starting from a clean semantic layer or navigating total chaos, this is your path to production-grade AI — no waiting required.