Generative AI for AI Developers, RPA Engineers, & Data Scientists

1.1 Fundamentals of Generative AI

• Definitions: discriminative vs generative, encoder/decoder/encoder‑decoder models
• Architecture deep dive: Transformer, attention layers, embeddings
• Pre-trained model survey: GPT-family, Claude, Titan, BERT-type

1.2 Prompt Engineering & RAG Pipelines

• Prompting techniques: zero-shot, few-shot, chain-of-thought (CoT)
• Reusable & dynamic prompting: templates, context injection
• Tokenization and prompt constraints
• Data chunking strategies and embedding generation
• Building RAG pipelines with Pinecone/OpenSearch
• ReRanking methods for relevance
• Hands-on:
  • Prompt refining in UI drag/drop or API playground
  • Vector-based document retrieval in Python (Pandas + LangChain)

1.3 Cost & Performance Tradeoffs

• Model hyperparameters: temperature, top-p, top-k, max tokens
• Token usage, request latency vs throughput
• Quality vs cost scaling strategies
• Tool: cost estimator and token budgeting tool (incl. LangSmith demo)

2.1 Pre-trained Model Access

• API vs SDK vs embedded container usage
• Role integrations:
  • AI Devs: code assistant with OpenAI’s API + Flask
  • RPA: Bedrock/UIPath connector drag/drop workflow
  • Data Scientists: summarization analytics with Python endpoint

2.2 Fine-Tuning & Transfer Learning

• Structured data prep, annotation pipelines
• Transfer learning with Hugging Face + SageMaker
• Live labs:
  • Train a domain-specific Q&A model
  • Fine-tune BERT on enterprise corpora

2.3 Evaluation Metrics & Monitoring

• Perplexity, BLEU, ROUGE, accuracy + fairness assessments
• Real-time performance monitoring (CloudWatch/SageMaker)
• Error logging, alerting, retraining triggers

2.4 Robustness, Bias & Security

• Adversarial testing, out-of-distribution detection
• Safety mechanisms: guardrails, content filters, prompt health scores
• Security best practices: prompt safety, injection prevention
• GDPR/Claude-access control discussions

3.1 Foundation Model Orchestration

• Amazon Bedrock, Lambda orchestration, Step Functions
• Build flows: API Gateway → Bedrock → data store
• Role labs:
  • Orchestrate GenAI flows in drag/drop
  • Bedrock-based microservice
  • Step Function orchestrated RAG pipeline

3.2 Agents, Vector Memory & Retrieval

• Setup of agent loops with retrieval context
• Tools: LangChain agents, AutoGen integration
• Agentic Frameworks: Crew AI, LangGraph deep dive
• Lab:
  • Build multi-tool agent summarizer or code scribe

3.3 Workflow Automation

• Combine SageMaker Pipelines with Step Functions
• Create reusable Python functions and UiPath custom activities
• Lab:
  • RPA system: multi-step intelligent bot
  • AI Dev: event-driven code assistant

3.4 API Rate Limiting & Scaling

• Throttling policies, retries, exponential back-off
• Cost optimization strategies with reserved capacity

4.1 Agentic Automation & Memory

• Agent architecture: task planning, memory, tool chains
• Memory management: vector and Redis session states
• Lab: build memory-enabled ticket summarization agent

4.2 Packaging & CI/CD Deployment

• Dockerize Python + model server
• Deploy via AWS Lambda, ECS, or UiPath Orchestrator
• CI/CD pipelines using GitHub Actions + CDK or UiPath pipelines
• Lab:
  • Package and deploy microservice
  • Versioned bot with rollout control

4.3 Governance, Ethics & Feedback Loops

• Implement user feedback pipelines (Ground Truth, custom forms)
• Ethical alignment: transparency, explainability & user trust
• Fairness checks, audit logs, consent flows, masked logging
• Risk management: data security, auditability

4.4 Final Project Presentation

• Build and demo end-to-end project per role:
• Code assistant microservice (AI Dev)
• Smart automated workflow (RPA)
• Interactive analytics agent dashboard (Data Scientist)
• Peer review and ethical checklist audit