Скачать или смотреть Build an API-Powered AI Travel Agent in Databricks

Most AI projects start with a complex architecture — but what if you could build an API-powered AI agent in minutes and reuse the framework for almost any use case?

In this episode of VectorLab, Merve Karali and Alexander Genser from Databricks show how they built a travel agent that can query real-time train schedules from the Swiss Railways API, powered by Databricks UC Functions, the Agent Framework, and MLflow 3.0.

This is an architecture can be reused for energy, finance, retail, or any scenario where you want an agent to interact with APIs, structured data, or unstructured data.

🧠 What You’ll Learn:
How to wrap any API in a Unity Catalog Function with access controls
How to build and test an agent in the Databricks Playground
How to export your prototype into production-ready code
How to evaluate and monitor agents with MLflow 3.0 (tracing, prompt registry, DSPy optimization)
How to deploy your agent as a Model Serving Endpoint with human review in the loop

📣 Call to Action:
🚀 Clone the GitHub repo and try it yourself →
https://community.databricks.com/t5/t...
https://github.com/databricks-solutio...
🔔 Subscribe to VectorLab for more practical GenAI + Databricks tutorials
💬 Comment if you have an API you’d love to turn into an AI-powered agent
👍 Like if you want more hands-on agent framework demos


🕐 Chapters / Key Moments
00:00 - Intro: AI-powered API travel agent in Databricks
01:01 - Why transportation? Why this use case?
02:10 - The challenge: simple architecture, broad applications
03:40 - Overview of the Swiss Railways API endpoints
04:56 - Who this is for: AI experts & non-AI specialists
06:43 - How the UC Functions + Playground workflow works
08:09 - Architecture breakdown: API to agent to deployment
10:33 - Wrapping APIs as Unity Catalog Functions
13:08 - Beyond APIs: structured & unstructured data retrieval
14:43 - Demo: building an agent in the Playground
17:01 - Testing train queries in real time
19:06 - Exporting to a driver notebook with MLflow 3.0
21:42 - Evaluating agents with MLflow: tracing & scoring
24:02 - Prompt registry & DSPy optimization
27:42 - Deploying the agent with Model Serving
29:14 - Human feedback loop for prompt and model improvement


#AI #GenAI #Databricks #AgentFramework #MLflow #UCFunctions #MachineLearning #AIinAction #APIAgent #TravelTech #SwissRailways #DataEngineering #LangChain #PromptEngineering #LLMs #OpenSourceAI #ModelServing #VectorLab #AIEvaluation #DSPy