16 Months of building
AI Systems.
How Can AI Agents
Help Businesses?
What are LLMs?
Understand human language and
generate text outputs.
ChatGPT, Google Bard, Anthropic Claude.
E.g. Q&A,
Summarizing Text,
Generating Code
Building and Using LLMs
Required extensive technical knowledge (hard code):
Write code to handle data preprocessing, training
loops, and evaluation, etc.
Integration challenges.
Introducing LangChain
Open Source framework (in JS and Python)
Enables chaining of models, tools and actions,
allowing LLMs to perform more complex tasks.
Reducing barriers for building real world AI
applications.
Abstraction Levels
Application-Level DocuSphere, Make, n8n, Instagram
Dev. /Deployment Tools Docker, Kubernetes, Github
APIs/ Services Google Maps API, Stripe API, etc.
Libraries & Frameworks LangChain, Django, React, TensorFlow
Programming Languages Python, JS, Java, C++
Machine Code/Assembly Binary, x86, ARM Assembly
OS Windows, Linux, MacOS
Hardware (CPUs, GPUs, Intel Core i9, NVIDIA GeForce RTX, etc.
Storage, etc.)
What can you build with
LLMs using LangChain?
Conversational Chatbots
Autonomous AI Agents
Data Analysis Systems
i.e. anything that involves language inputs.
LangChain Basics- Chains
Chains are sequences of operations or steps
where each step is either a function, a model (e.g.,
an LLM), or an action that depends on the result
of the previous step.
Concept of Chains in No-Code
LangChain Basics- Agents
An LLM with a Tool
Can make decisions; act as reasoning engines.
Interact with Tools.
Autonomous, doensn’t need constant human
input.
As a consequence, it is able to autocorrect itself when
any errors appear (if trained well).
Langchain Agents- Example
Show me a graph of
the performance of
Apple’s Stock from
the last 6 months
Langchain Agents- Example
Hmm.. let me check the web API for
the current date and time.
Show me a graph of
the performance of
Apple’s Stock from
the last 6 months
Langchain Agents- Example
Alright, so today is September 7th
2024. Lets look for data from March
to September.
Show me a graph of
the performance of
Apple’s Stock from
the last 6 months
Langchain Agents- Example
Great, now I have the stock price
data. Now I will generate a graph
for this.
Show me a graph of
the performance of
Apple’s Stock from
the last 6 months
“As of m
ytraining
data, a
RAG Systems
know i ll I
s X”
Technique to enhance Large Language Models (LLMs) by
connecting them to knowledge sources.
Addresses limitations of LLMs by allowing them to reference
knowledge bases outside their training data.
Store information in a vector store and then retrieve relevant
information to the user.
Combines the following elements:
Embedding Models
Vector Stores
LLMs
You might have used RAG already!
(Knowledge base modules in
VoiceFlow, BotPress, etc.)
Vector Spaces
The axes in a vector store's vector space
represent latent, learned features of the
data, not concrete or manually defined
properties.
Each point in the vector space
corresponds to an embedding vector
of fixed dimensionality.
The position of vectors along these axes
determines their distance from one
another, which is used to measure
similarity between items.
RAG≠ AI Agents
RAG systems only retrieve additional
knowledge.
Not autonomous
Cannot solve complex problems
Web Search
Python IDE
Web Search
RAG
Python IDE
Multi-Agentic Systems
CrewAI
ChatDev
Microsoft AutoGen
Multi-Agentic Systems
Chain of Smaller Specialised Models >>
As long as the tasks are not highly
interdependent.
AI Agents vs RAG
While RAG uses an LLM and a retrieval system (a type of tool), it is not
typically classified as an agent because it lacks the decision-making and
dynamic action-taking abilities of a true agentic system.
RAG focuses on improving text generation with external information,
whereas agents are designed to make autonomous decisions and
perform tasks using a variety of tools, beyond just generating text.
AI Agent vs AI Automation
AI Automation:
Follows fixed rules to complete specific, repetitive tasks.
Reacts based on predefined workflows.
AI Agents:
Can make decisions and take actions independently.
Flexible, adapts to new situations and learns from interactions.
Why are Agents a big deal?
Handle Complex Human Tasks.
Mimic Humans in text-based tasks.
Speed up processes that traditionally require manual effort
(e.g. automating customer service, lead gen.)
Enable human focus on creative and strategic roles.
Building Your First Agent
AI Agent ≠ AI Automation!
Must I know how to code extensive
Python, use IDEs, know Front End
code, host on servers, etc. ?
No!
You can start here:
Low-Code
Why is n8n good for
learning ai agents?
n8n is great for you to understand langchain
and ai agent components without having you to
delve deep into programming.
You stay a couple of abstraction layers away.
All of the logic can be handled inside n8n.
Demo Time!
Airbnb Chatbot on
n8n
AI No-Code Tools
Conversational AI Chatbots:
Voiceflow
Chatbotbuilder
BotPress
AI Automations Voice Agents
Make No-Code Bland.ai
Zapier No-Code VAPI
Synthflow
Adoption in Enterprise
Deloitte has integrated AI-driven tools for automating complex financial
data analysis and reporting.
JP Morgan has utilized AI agents to automate its email marketing. These
agents are not only sending targeted emails but also analyzing feedback
to refine and perfect their messaging.
Lenovo has implemented AI agents in two key areas: software
engineering and customer support.
Adoption in Enterprise
We are here ;)
Considerations when
building quality solutions
Data is KING
Prompt Engineering
System Prompt
Role Objective Tools
Detailed & CONCISE Instructions
Output & Format
Examples!
Additional Notes/Reminders (tweaking)
Other useful things to know
Integrations & APIs
Method 1 : API Calls
Request Method
GET, POST
Request URL {
"name": "John Doe",
Request Body "email":
"john@example.com"
}
Request Headers Authorization: Bearer <token>
Response Body
Method 2 Easiest: Webhooks
Webhooks: webhook URL
Request Body
Only POST requests
Learn the basics of Python and
Javscript
Variables and Data Types
Conditional Statements
Lists and Arrays
Parsing Data
Authentication Types
OAuth 2.0: A more advanced authentication method that
uses tokens to provide access to resources on behalf of a
user.
Access Token: Generated after user consent and
used to authenticate requests.
Refresh Token: Used to obtain a new access token
without re-authentication.
Understand the Clients
Needs and Objectives
Effectively
Creating Draft
Architectures
Agents are the Future.
Agents will:
Handle monotonous specific tasks;
answer questions, etc.
Tasks that connect to software and use
text.
Voice solutions
Human Employees will work on more abstract
tasks:
PR
Meetings and Deals
etc.
Thank You!
