cropped image_6_ removebg preview
BECOME AN INSTRUCTOR
shape
shape

Generative AI with Databricks

  • Home
  • Generative AI with Databricks

Generative AI with Databricks

Duration: 32 Hours

Course Overview

  • This course provides a comprehensive understanding of building, deploying, evaluating, and monitoring Generative AI solutions using Databricks. The course covers key concepts such as Retrieval-Augmented Generation (RAG), LLM applications, prompt engineering, data preparation, deployment, and evaluation techniques.

Target Audience

  • Machine Learning Practitioners, Data Engineers, AI Solution Architects, and Technical Leaders interested in developing Generative AI applications using Databricks.

Pre-requisites

  • Familiarity with Natural Language Processing (NLP) concepts
  • Knowledge of prompt engineering best practices
  • Basic understanding of the Databricks Data Intelligence Platform
  • Familiarity with concepts like vector databases, embeddings, and RAG architecture

Course Outline

Module 1: Generative AI Basics

  • What is Generative AI?
  • Generative Models
  • Generative AI Use Cases
  • LLMs and Generative AI
  • LLMs are not hype—they change the AI game
  • What is an LLM?
  • How do LLMs work?
  • LLMs generate outputs for NLP tasks
  • LLM business use cases

Module 2: LLMs and Generative AI

  • LLM Applications
  • LLM Flavors
  • Using Proprietary Models (LLMs-as-a-Service)
  • Databricks AI
  • Databricks Data Intelligence Platform
  • Building Gen AI applications on Databricks
  • Databricks AI — a data-centric AI platform
  • Databricks AI — optimized for Generative AI
  • AI Adoption Preparation
  • How to Prepare for AI Revolution
  • Strategic Roadmap for AI Adoption

Module 3: Legal and Ethical Considerations

  • Potential Risks and Challenges
  • Legal Issues
  • Ethical Issues
  • Social/Environmental Issues
  • Legal Considerations
  • Data Privacy in Generative AI
  • Data Security in Generative AI
  • Intellectual Property Protection
  • Litigation and/or other regulatory risks
  • Ethical Considerations
  • Fairness and Bias in Data
  • Bias Reinforcement Loop
  • Reliability and Accuracy of AI Systems
  • Auditing Generative AI Models
  • Human-AI Interactions
  • How will AI Impact Society
  • AI and Workforce

Module 4: From Prompt Engineering to RAG

  • Prompt Engineering Primer
  • What is prompt engineering
  • Elements of a prompt (instruction, context, input data, output format)
  • Basic prompt engineering techniques
  • Zero-shot learning
  • Few-shot learning
  • In-context learning
  • Prompt-chaining / Chain-of-Thought prompting
  • Best practices for prompt engineering
  • Formatting tips
  • Introduction to RAG
  • How language models learn (pre-training, fine-tuning, contextual info)
  • Use vectors to improve factual recall
  • Define RAG and its role in closing knowledge gaps
  • RAG applications (Q&A chatbots, search augmentation, content creation)
  • Key components of RAG (vector store, index & embed, retrieval)
  • Using Databricks features for RAG workflows
  • Demo: In-Context Learning with AI Playground
  • Prompt which hallucinates vs. prompt which doesn’t hallucinate
  • Augment prompts with additional context

Module 5: Preparing Data for RAG Solutions

  • Data Preparation for RAG
  • Importance of data prep for RAG
  • Chunk, embed, and store data effectively
  • Utilize Delta Lake for reliable analytics
  • Apply chunking strategies (fixed-size, content-aware)
  • Demo: Preparing Data for RAG
  • Extract PDF content as text chunks
  • Create embeddings with foundation model API
  • Save embeddings to a Delta table
  • Lab: Preparing Data for RAG

Module 6: Mosaic AI Vector Search

  • Introduction to Vector Stores
  • Overview of vector databases for high-dimensional vectors
  • Role in RAG architecture
  • Use cases (recommendation engines, similarity search)
  • PQ technique and similarity metrics
  • Mosaic AI Vector Search
  • Integrated platform for storing vector data with metadata
  • Vector search workflows and methods
  • Create vector search endpoints
  • Demo: Create Vector Search Index
  • Data preparation and chunking
  • Create and configure vector search index
  • Perform similarity search and re-rank results

Module 7: Assembling and Evaluating a RAG Application

  • Assembling a RAG Application
  • Workflow phases (development, testing, evaluation, production)
  • MLflow for monitoring and logging
  • Model packaging with MLflow
  • Using MLflow Model Registry for model management
  • Evaluating a RAG Application and Continual Learning
  • Evaluate components (chunking, retrieval, generation)
  • Use metrics like precision, relevancy, and recall
  • Assess answer correctness for generator performance
  • Demo: Assembling and Evaluating a RAG Application
  • Set up RAG components (retriever, model, pipeline)
  • Save model to Unity Catalog
  • Lab: Assembling a RAG Application

Module 8: Compound AI Systems

  • Defining Compound AI Systems
  • What are compound AI systems?
  • Examples (RAG, agent-based chains)
  • Managing multiple intents with AI systems
  • Designing Compound AI Systems
  • System architecture and workflow
  • Identify and classify intents
  • Choose tools for tasks (web search, API interaction)
  • Demo: Deconstruct and Plan a Use Case
  • Application planning
  • Multi-endpoint architecture

Module 9: Model Deployment Fundamentals

  • Model Management
  • Packaging forms for LLMs
  • Workflow management
  • Model lifecycle (versioning, deployment options)
  • Deployment Methods
  • Batch, streaming, real-time, and embedded deployment
  • Latency and throughput analysis

Module 10: Batch Deployment

  • Introduction to Batch Deployment
  • Ideal use cases for batch processing
  • Querying foundation model API for batch inference
  • Scaling batch workloads
  • Demo: Batch Inference
  • Model registration to Model Registry
  • Single-node and multi-node batch inference
  • Using ai_query() for batch inference
  • Lab: Batch Inference Workflow

Module 11: Real-time Deployment

  • Introduction to Real-time Deployment
  • Serving ML models in production
  • Low-latency applications
  • Databricks Model Serving
  • Managing models through a single platform
  • Deploying models behind robust APIs
  • Demo: Deploying an LLM Chain to Model Serving
  • Reload chain/model
  • Deploy chain to a serving endpoint
  • Explore UI and metrics

Module 12: AI System Monitoring

  • AI Application Monitoring
  • Proactive approach to maintaining performance
  • Accuracy, resource usage, data quality
  • Using Lakehouse architecture for monitoring
  • Demo: Online Monitoring an LLM RAG Chain
  • Unpack inference table
  • Create SQL alerts and dashboards
  • Lab: Online Monitoring

Module 13: Gen AI Evaluation Techniques

  • Evaluation Techniques
  • Compare LLM evaluation with traditional ML evaluation
  • Task-specific evaluation techniques (BLEU, ROUGE)
  • LLM-as-a-Judge approach
  • Demo: Benchmark Evaluation
  • Setup for text summarization
  • Compute ROUGE metric
  • Lab: Domain-Specific Evaluation

Module 14: End-to-End Application Evaluation

  • End-to-End Evaluation
  • Evaluate entire AI systems for performance and cost
  • Multi-component architecture evaluation
  • Use custom metrics for system goals

T. Sanjay

Tech Enthusiast | Seasoned Corporate EnterT(r)ainer

Apply Now