Build and Maintain your Company Copilot with Azure ML and GPT-4 Q&A
summary
Shows how to assemble LangChain or Semantic Kernel code into Prompt Flow UI and SDK for side‑by‑side evaluation, tracing each nested LLM call with a tool decorator, and running comparisons across ~10 built‑in metrics. Runs evaluation jobs on a per‑user compute instance, packages deployments as containers to an endpoint, and stores credentials in Key Vault. Recommends exhausting prompt‑engineering before model tuning, uses Ada fine‑tuning for classifiers where cheap labeled gains appear, and leverages 32K token context and chained prompts for large retrieval workflows.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 1:12Company Copilot with prompt frameworks
- 3:23Prompt flow evaluation workflow
- 5:54Compute environments in prompt flow
- 8:22Data sources and prompt complexity
- 10:51Multi-cloud support
- 13:00LLM creative uses
- 15:06AI data security
- 18:03LangChain vs Semantic Kernel
- 19:29LLM app performance
- 21:36Prompt engineering future
- 25:11Detecting prompt drift
- 26:17CI/CD for prompt flow
- 27:39Fine-tuning vs prompt engineering
- 30:32RAG data contamination
- 33:01AI advancements future
- 37:25Training models with facts
- 40:19Scaling LLM plugins
- 42:49Closing remarks
related talks
01h 58mBuild, Evaluate and Deploy a RAG-Based Retail Copilot with Azure AI
42mCode First LLMOps With Azure AI & Prompty
01h 08mIn Prompts We Trust
45mBuilding a copilot: Azure AI Studio or Copilot Studio
40mCode-First LLMOps from prototype to production with GenAI tools
26mMastering Large Language Models: Prompt Engineering for Precision & Performance
44mExploring ChatGPT for Improved Observability
01h 00mUnleashGPT: Mastering the Implementation of Large Language Models