营销AI代理：利用Mosaic AI创建多用途代理营销助理.pdf

1、AI Agents for MarketingUsing Mosaic AI and LangGraph to create a Multi-Purpose Agentic Marketing AssistantAuthor NameSailesh Bharathwaaj KrishnamurthyAgenda3IntroductionIntro to 7-ElevenEarly ChatbotsHow we started building chatbotsMarketing Assistant ArchitectureProblem statement and the agentic ar

2、chitectureExample WorkflowWalkthrough of an example scenarioImplementation detailsLangGraph and Mosaic AIResults&Next stepsWhat next?Introduction to 7-Eleven 50%of US population within 2 miles of a 7-Eleven7-Eleven has a worldwide presence!First to offer coffee on-the-goFirst self-serve soda fountai

3、nFirst convenience store to offer gas7Now Delivery across US&CanadaEstablished in 1927;a pioneer in convenienceOver 13,000 Stores in US&Canada85,000 Stores Globally in 19 Countries&RegionsIntroduction to 7-Eleven 57-Eleven also has a History of Tech InnovationCashierless StoreAI self-checkoutThe Cha

4、tbot GraveyardEarly ChatbotsEarly ChatbotsBackground:Utilized Falcon 7B&quantized Llama-1 modelsVery naive RAG patterns and basic prompt tuning.Availability was an issue for newer modelsKey Learnings:Importance of document pre-processing for retrieval bots.Effective chunking strategy&summarizationMu

5、lti-step reasoning;an early version of agentic models7Weve been building Chatbots since the first Open-source ModelsEarly ChatbotsSpeaker&Session info.in two different UC tablesCreated a Genie Space with clear comments&detailed promptsExposed the Genie Space as an API&integrated it with Microsoft Te

6、ams 8We Built a Teams Chatbot for Our Own Data&AI Summit(Oct.24)Early ChatbotsUsed multi-step reasoning to make sure the answers were always relevant.This framework was extended to other chatbot use cases as well.9We Battle Tested a Production Architecture for Around 500 UsersEarly Chatbots10Guardra