Gen AI Syllabus

Welcome to “Generative AI with LangChain ” course…

A Conceptual Course Syllabus

Course Overview

This course introduces how modern AI assistants are built, from the basic idea of talking to an AI, all the way to creating smart helpers that can remember conversations and take actions on your behalf. The focus is on concepts and understanding — no deep coding background required to follow along.

What Generative AI Really Is

Unit 1 — Getting Started with Generative AI

• The difference between AI that classifies things and AI that creates things
• The kinds of content AI can generate today: text, images, audio, code
• Where large language models fit in the bigger picture

Understanding Language Models

• How language models learn from huge amounts of text
• What limits a model’s memory in a single conversation
• The difference between training a model and simply guiding it with instructions
• Open vs. paid models — trade-offs at a glance

Controlling How the AI Responds

• Making answers more predictable or more creative
• Limiting response length
• The balance between consistency and originality

Why LangChain Exists

• The problem it solves for anyone building with AI
• The main building blocks you will meet again and again
• How these blocks connect to form real applications

Unit 2 — Talking to AI Models

Two Ways to Ask an AI a Question

• The simple “send text, get text back” style
• The conversation style with roles (system, user, assistant)
• When to prefer one over the other

The Roles in a Conversation

• The system voice that sets the AI’s personality and rules
• The user voice that asks questions
• The AI voice that responds
• Why the order of messages matters

Switching Between AI Providers

• The idea of a common interface across different AI companies
• Choosing between models running online vs. on your own machine

Practical Concerns When Working with AI Services

• Keeping your access keys safe
• Working behind corporate networks
• Being mindful of cost, speed, and usage limits

Unit 3 — Writing Better Instructions (Prompt Design)

The Basics of Good Prompting

• Asking without examples vs. giving a few examples
• Guiding the AI to “think step by step”
• Being clear about the tone and format you want

Reusable Instruction Templates

• The idea of a fill-in-the-blank prompt
• Feeding different inputs into the same template

Conversation-Style Templates

• Combining a fixed personality with changing user questions
• Building chat-based experiences on top of templates

Remembering the Conversation So Far

• How to feed past turns back into the AI
• Handling long conversations that outgrow the AI’s memory

Managing Prompts Like Any Other Asset

• Saving prompts to a file so anyone on the team can edit them
• Versioning and reusing prompts across projects

Unit 4 — Getting Clean, Predictable Answers

Why Free-Form Text Is a Problem

• The pain of parsing paragraphs to find the answer
• The need for consistent, ready-to-use results

Describing the Shape of the Answer You Want

• Simple ways to lay out the fields you expect
• Adding rules such as “must be a number between 1 and 10”
• The concept of a schema — a contract for the AI’s reply

Making the AI Follow Your Shape

• Attaching your expected shape to the model
• What happens behind the scenes to enforce it

Working with Different Kinds of Models

• Models that natively support structured answers
• Models where you have to coax the format through instructions

Handling Things Going Wrong

• Retrying when the answer doesn’t fit
• Sensible fallbacks when the AI misbehaves

Unit 5 — Turning AI Answers into Usable Data

The Job of an Output Parser

• Cleaning up the raw response into something you can work with

Common Parsing Styles

• Plain text extraction
• Structured data extraction (lists, dictionaries, records)
• Validated objects that follow your rules

Telling the AI How to Format Its Reply

• Automatically including the format instructions in the prompt

Choosing Between Parsing and Enforced Structure

• When each approach shines

Unit 6 — Chaining Steps Together

The Idea of a Pipeline

• Every step is a Lego block you can snap onto the next
• Running steps one after the other, or all at once

Common Pipeline Patterns

• A single step: ask a question and get an answer
• A relay race: the answer of one step becomes the question for the next
• Teamwork: several steps run in parallel and their results are combined
• Decision points: pick a different path based on what the AI said

Seeing Inside Your Pipeline

• Visualising the flow of a chain
• Turning on detailed logs to see what happened at each step
• Introduction to tracing tools for deeper visibility

Unit 7 — Giving the AI Real-World Abilities (Tools)

Why the AI Needs Help

• The AI is great at thinking and writing, but weak at doing
• Examples: fetching live prices, running calculations, searching the web, sending an email

What Counts as a Tool

• A name, a description, and a set of inputs
• Why the description matters — the AI reads it to decide whether to use the tool

Ready-Made Tools You Can Plug In

• Web search
• Running commands on your computer
• Encyclopedia lookups, math engines, and more

Building Your Own Tools

• Three levels of effort: quick, medium, and full control
• Deciding when a simple tool is enough vs. a fully customised one

Grouping Related Tools Together

• The idea of a toolkit — everything an assistant needs for a domain

Unit 8 — How the AI Decides to Use a Tool

The Three-Step Dance

• Step 1 — Introducing the tools to the AI
• Step 2 — The AI proposes a tool to use and what inputs to give it
• Step 3 — Your code actually runs the tool and hands back the result

Key Insight — the AI Does Not Actually Run the Tool

• The AI only decides; your program does the work
• This separation keeps things safe and predictable

Handling Multiple Tools in One Turn

• Letting the AI call several tools at once

Values the AI Should Not See

• Passing behind-the-scenes information (like a user ID or a live rate) without exposing it

Unit 9 — Agents (AI That Takes Action on Its Own)

What Makes Something an Agent

• The AI keeps thinking and acting in a loop until the task is done
• The reason–act–observe pattern

Building Your First Agent

• Combining a model, a set of tools, and a personality
• How you talk to it and what it gives back

Common Agent Types

• Agents with a single ability
• Agents that pick from many abilities
• Agents that return neatly structured final answers

Shaping the Agent’s Behaviour

• Using a system message to set the tone and scope
• Restricting what the agent is allowed to do

Keeping Agents Trustworthy

• Preventing endless loops
• Recovering gracefully when a tool fails
• Avoiding made-up tool names or bad inputs

Unit 10 — Memory and Ongoing Conversations

Why Memory Matters

• By default, the AI forgets everything after each reply
• Memory is what makes an assistant feel personal

Short-Term Memory

• Passing recent messages back into every request
• Trimming or summarising when history gets too long

Long-Term, Persistent Memory

• Saving conversation state so it survives across sessions
• The idea of a conversation ID that identifies each user or thread

Keeping Users Separate

• Making sure Alice’s notes never leak into Bob’s session
• Isolating each user’s data cleanly

Beyond Chat History

• Storing facts in a database or knowledge store
• A gentle introduction to retrieval-based memory

Unit 11 — A Look at More Advanced Workflows

When a Straight Line Isn’t Enough

• Simple pipelines can’t handle branching, loops, or human checkpoints
• The idea of representing a workflow as a network of steps

Workflows That Remember Where They Left Off

• Pausing, resuming, and checkpointing complex processes

When to Reach for Advanced Orchestration

• Human-in-the-loop reviews
• Multiple specialised agents cooperating on one task
• Custom decision logic that goes beyond a simple pipeline

Unit 12 — Building Real Applications

Adding a User Interface

• Quickly turning your AI logic into a working web page
• Wiring form inputs into your prompts and agents

Common Application Patterns

• Question-and-answer tools and summarisers
• Content generators (study notes, quizzes, marketing copy)
• Task-oriented assistants that remember and act

Organising a Real Project

• A sensible folder structure that separates concerns
• Keeping settings and secrets outside your code
• Managing dependencies so anyone can run your project

Unit 13 — Making It Ready for the Real World

Seeing What’s Happening Inside

• Logs, tracing, and dashboards
• Tracking which tools were used, how long things took, and what it cost

Measuring Quality

• Building a set of “golden” examples to test against
• Using AI itself to help judge answer quality

Safety and Guardrails

• Checking what goes in and what comes out
• Awareness of prompt manipulation attempts
• Protecting personal data and limiting what tools can touch

Cost and Performance

• Using cheaper models where they’re good enough
• Caching repeated work, streaming answers, running things in parallel

Guiding Principles Throughout the Course

Principle	What It Means
Build Small, Combine Freely	Every piece is designed to snap together. Start with tiny building blocks.
Prefer Structured Data Over Loose Text	Whenever data crosses a boundary, give it a clear shape.
Separate the Layers	Keep the instructions, the model, the tools, and the workflow independent.
Let the AI Reason, Let Code Do the Rest	Use the AI only where judgement is needed. Everything else belongs in normal code.
Make the Invisible Visible	If you can’t see what your assistant did, you can’t improve it or trust it.

Recommended Learning Path

Follow the units in order — each one builds on the previous. Skipping ahead (especially past Tools and How the AI Uses Tools) tends to make Agents feel like magic.

Production Readiness