GenAI and Agentic AI BootCamp

Advanced Prompting Patterns

• Chain-of-Thought (CoT), Tree-of-Thought (ToT), ReAct, self-consistency, prompt chaining, few-shot design

Prompt Debugging & Benchmarking

• Hallucination detection, A/B testing prompts, log probability analysis, reliability scoring, prompt benchmarking

Context Engineering ★

• What to put in the context window — compression, pruning, routing, summarisation, context budgets

Structured Outputs ★

• JSON mode, Pydantic v2 validation, function calling schemas, output parsers, retry on parse fail

Prompt Versioning

• Managing prompts as code — versioning, templates, LangSmith prompt hub AI-

Native Dev Tools ★

• Cursor + Claude Code as development accelerators — used throughout the bootcamp from this module forward

Embeddings & Similarity

• Metrics Embedding model comparison (OpenAI, Cohere, BGE), cosine/dot product/Euclidean metrics, dimensionality tradeoffs

Chunking Strategies

• Semantic chunking, sliding window, recursive splitting, metadata tagging, parent-child chunks

Vector Databases

• FAISS, Pinecone, Weaviate, Chroma, pgvector — selection criteria, index management, production operations

Advanced Retrieval

• Hybrid search (dense + BM25), reranking (Cohere/BGE), query rewriting, HyDE, multi-query retrieval

Agentic RAG ★

• RAG inside an agent loop — agent decides WHEN to retrieve, WHAT to retrieve, iterative/self-correcting retrieval, routing between strategies

RAG Evaluation

• RAGAS framework, context precision/recall, faithfulness, answer relevance, LangSmith eval datasets

Multi-Document Reasoning

• Cross-document retrieval, citation chains, conflicting source resolution

Parameter-Efficient Fine-Tuning

• LoRA, QLoRA, PEFT — when to fine-tune vs RAG vs prompt engineer, cost-benefit analysis

Dataset Preparation

• Instruction datasets, ShareGPT/Alpaca format, data cleaning, quality filtering with Pandas

Model Evaluation

• BLEU, ROUGE, perplexity, domain-specific evals, model versioning with Weights & Biases, Matplotlib visualisation

Vision-Language Models

• GPT-4o vision, Claude 3.5 vision, multimodal embeddings, image + text reasoning chains, multimodal reasoning

Diffusion Models ✚

• How diffusion works — noise schedules, denoising process, Stable Diffusion, DALL-E, text-to-image APIs, use in agent pipelines

Document AI / ADE ★

• Agentic Document Extraction — PDFs, scanned docs, tables, handwriting, mixed-language content. Unstructured.io, LlamaParse

GenAI to Agentic Evolution

• From chatbots to agents — autonomy levels, decision loops, agent vs workflow vs pipeline distinction

Agent Components

Planner, Executor, Tools, Memory — roles, interactions, failure modes of each component

Architecture Patterns

• ReAct, Plan-and-Execute, Tree of Thought, Reflection loops, self-critique patterns

Goal Decomposition & Task Queues ✚

• Breaking high-level goals into subtasks, task queue management, dependency resolution, execution ordering

Decision Making Under Constraints ✚

• Agents operating within budget limits, time constraints, tool availability — constraint-aware planning

Agent Types

• Reactive vs deliberative vs hybrid agents, goal-driven vs utility-based, single vs multi-agent contexts

LangGraph Deep Dive

• Stateful graph design, cyclic graphs, node types, conditional edges, checkpointing, state reducers

CrewAI Framework

• Role-based agents, task delegation, crew orchestration, process types (sequential/hierarchical)

AutoGen / AG2

• Conversational agents, group chat managers, code execution agents, agent negotiation patterns

OpenAI Agents SDK ★

• Native OpenAI approach — agents, handoffs, guardrails, tracing. Rapidly becoming enterprise standard in 2026

Framework Tradeoffs

• LangGraph vs CrewAI vs AutoGen vs OpenAI SDK — complexity, flexibility, enterprise readiness, community size

Semantic Kernel

• Microsoft SK for enterprise orchestration, plugin design, Azure AI Foundry integration

Function Calling & Tool Schemas

• OpenAI/Claude function calling, tool schema design, Pydantic models, output validation, parallel tool calls

Tool Integration

• REST/GraphQL API calling, browser tools, file system read/write, code execution sandboxes

Short-Term Memory

• Conversational context, window management, token-aware summarisation buffers

Long-Term / Vector Memory

• FAISS/Pinecone/ChromaDB for persistent memory, episodic vs semantic vs procedural memory

Memory Architecture Patterns ★

• mem0, Zep, MemGPT — purpose-built agent memory systems, choosing the right architecture per use case

Memory Persistence

• LangGraph memory nodes, cross-session state, memory retrieval and update strategies

Multi-Agent Orchestration

• Role assignment, task decomposition, agent handoffs, parallel vs sequential execution, supervisor patterns

Message Passing & Negotiation

• Inter-agent communication, shared state management, conflict resolution, consensus mechanisms

Human-in-the-Loop (HITL) ★

• Approval gates, escalation paths, oversight checkpoints for healthcare/finance/legal — fail-safe patterns

MCP — Model Context Protocol

• Architecture, tool/resource exposure, server/client design, interoperability with external services

A2A & ACP Protocols

• Agent-to-Agent protocol, Agent Collaboration Protocol — comparison, when to use each, implementation with CrewAI/AutoGen

Evals Framework ★

• Defining eval metrics, building eval pipelines, regression testing agent behaviour — RAGAS, Braintrust, UpTrain

LangSmith Deep Dive

• Tracing, evaluation datasets, prompt versioning, A/B testing, production monitoring dashboards

Agent Monitoring & LLMOps

• Latency logging, retry mechanisms, fallback logic, cost/token tracking, alerting on failures

Safety & Guardrails

• Prompt injection defense, loop prevention, constraint design, goal alignment, Guardrails AI / NeMo Guardrails

RBAC & Compliance

• Permission scoping, API key lifecycle, audit logging, trust boundaries, GDPR/HIPAA/SOC2 considerations

Containerisation & CI/CD

• Dockerising agents, GitHub Actions pipelines, environment variables, secrets management

Cloud Deployment Options

• HuggingFace Spaces, AWS Lambda, GCP Cloud Run — selection criteria, cost comparison, scaling strategies

Azure AI Foundry

• Azure AI Studio setup, Azure OpenAI Service, Azure AI Search for enterprise RAG pipelines, agent development on Azure

Semantic Kernel on Azure

• Orchestrating agents with SK, connecting to enterprise business data, Copilot Stack integration

Azure Monitoring & Performance ★ 

• Azure Monitor, Application Insights for agents, logging agent decisions, performance optimisation on

Azure Azure Cost Management ★

• Using Azure free credits efficiently, avoiding unnecessary API costs, cost alerts, right-sizing deployments

Microsoft Applied Skills Badge

• Guided MS Learn lab completion, step-by-step instructor guidance, official MS Applied Skills badge from Microsoft

SaaS Pricing Models

• Per-call, subscription, usage-based — token budgeting, cost modelling, margin calculation for production agents

Prompt Caching in Production

• GPTCache, Anthropic/OpenAI caching APIs — 60–80% cost reduction for repeated context in agent loops

Scheduling & Persistent Jobs ★

• Running agents on a schedule (cron-style), background job management, queue systems (Celery/Redis), agent status dashboards

Model Routing & Cost

• Optimisation Routing between cheap/expensive models dynamically, semantic caching, budget guardrails per agent

Agent Productisation

• Multi-tenant architecture, API gateway design, rate limiting, packaging agents as commercial products

Streamlit / Gradio

• UIs Building interactive demo UIs for rapid prototyping, client demos, and SaaS MVPs

AI Coding Agent Landscape

• GitHub Copilot Workspace, Devin-style agents, Claude Code, Cursor Agent mode — what enterprises are deploying in 2026

Agents that Write & Test Code

• Using LLM agents to generate, refactor, debug and test code — the #1 enterprise use case in 2026

Building a Coding Agent

• Hands-on: build a simple code-generation + test-execution agent using OpenAI Agents SDK or LangGraph