Business & Discovery

AI Foundations

• The AI landscape: traditional ML vs. Generative AI vs. Agentic AI

• How LLMs work — tokens, context window, inference

• Capabilities and limits: hallucination, knowledge cutoff, reasoning limits

• The cost / latency / quality triangle

• The AI Leader role — responsibilities, interfaces, and boundaries with Data Science and IT

• The AI market: providers, models, open vs. closed

AI Strategy & Use Case Prioritization

• Prioritization frameworks: value vs. feasibility, effort vs. impact

• Building a business case: ROI, TCO, build vs. buy

• Organizational AI maturity models

• Change management and AI adoption across the organization

• Governance and risk: bias, security, privacy, IP

• Communicating AI to leadership — framing and managing expectations

Business Discovery & Workflow Definition

• Stakeholder interview techniques — the right questions, surfacing pain points

• Process mapping: identifying automation and augmentation candidates

• Distinguishing good AI use cases from poor ones

• Requirements gathering for AI projects — how it differs from traditional IT

• Workflow design: task decomposition, human-in-the-loop

• Defining success metrics and acceptance criteria

Project Definition & Statement of Work (SOW)

• SOW structure for an AI project: scope, deliverables, assumptions, exclusions

• Defining measurable KPIs and success metrics

• Defining MVP vs. pilot vs. production

• Milestones and decision points in an AI project

• Managing uncertainty and experimentation in AI projects

• Cost structure: tokens, compute, development, maintenance

• Interfacing with technical teams — a shared language

Data Readiness for AI

• Data readiness assessment: is the data present, accessible, and good enough?

• Working with data and IT teams — asking the right questions

• Types of data sources: files, DBs, APIs, unstructured documents

• Data quality, lineage, and governance

• Embeddings and vector databases — concept and when to use them

• Preparing data for RAG: chunking, indexing, metadata

• Privacy, security, and access to sensitive data

Discovery to SOW Workshop

• Consolidating and weaving together all Part A tools

• Structuring a business-case presentation to a panel

• Preparing for the technical part: what the SOW tells us to build

Technical Build

LLMs & Prompt Engineering

• High-level LLM architecture — tokens, context, temperature

• Model types: closed vs. open, and the major providers

• Working with an LLM API in Python: request, parameters, error handling

• Prompt design: instruction, context, examples, output format

• Structured outputs and JSON mode

• Cost management and token budgeting

• Evaluating response quality: building test sets

Knowledge Bases & RAG Implementation

• Why AI systems need a Knowledge Base

• Extracting  text from PDF and Word with a dedicated Python library

• Text processing: chunking strategies and best practices

• Embedding models — what they are and how to use them

• Vector databases — concept and implementation

• Building a RAG pipeline with LangChain

• Connecting the KB to the LLM for document-grounded answers

• Reducing hallucinations and grounding responses

Tools & MCP Integration

• What MCP (Model Context Protocol) is and why it exists

• Installing and configuring a FastMCP environment

• Writing a simple tool in Python with FastMCP

• How an LLM decides when to call a tool and what to send

• The agent loop: plan → tool call → observe → respond

• Agent-as-a-tool:using one agent as a tool of another

• Connecting to the AI Platform: exposing an MCP server to the interface

Agentic Workflows & Multi-Agent Systems

• The difference between a bot and an agent

• Agent architectures: ReAct, Plan-and-Execute

• Memory and state management in agents

• Multi-agent systems: orchestration and specialization

• Reliability: guardrails, error recovery, human approval

• Testing and debugging agents: tracing, observing decisions

• Security: prompt injection, tool misuse, scope limitation

• Code review of agents — common pitfalls

AI Flow Builders & Automation

• AI data flows and pipelines — AI-oriented ETL

• Designing conversational workflows (Chat-as-a-Flow)

• AI Platform: building a visual flow, connecting nodes

• AI Platform: automating business processes

• Connecting multiple agents in a single flow

• Error handling and fallback paths in a flow

• Automating business processes with AI

Analysis, Debugging & Optimization

• Testing and validation of AI systems

• Identifying bottlenecks — latency, cost, quality

• Improving accuracy and response quality

• Performance optimization: caching, batching

• Monitoring: logging, tracing, dashboards

• An iterative improvement process

• When the solution is ready for a pilot

Pilot Deployment & Stakeholder Presentation

• Deployment patterns: from POC to pilot to production

• Managing stakeholder expectations at the pilot stage

• Defining success criteria for a pilot

• Feedback loops: collection, analysis, and iteration

• Preparing a demo for a non-technical audience

• Governance in a pilot: who is accountable, what is measured

• What happens after the pilot — a continuation plan

Final Project — End-to-End AI Solution

Deliverables

• A complete, updated SOW — based on Module 4 and refined after Part B

• Knowledge Base — relevant documents for the use case, indexed

• A working agent — at least 2 functional tools, connected to the KB

• AI Platform — a flow that presents the use case end-to-end

• Demo— a presentation to a panel: business context, live demo, Q&A