Ch 1 — Tech Stack Overview

What each tool does, how they connect, and the full-stack picture for Agentic AI
Index Under the Hood →
High Level
-
Click play or press Space to begin...
Step - / 8
A The Big Picture What makes AI "agentic"
1
person
User Sends a goal
or question
goal
psychology
LLM Reasons about
what to do
decides
build
Tools Search, APIs,
databases, code
result
refresh
Iterate Loop until
goal is met
2
arrow_downward Between user and LLM sits the orchestration layer
B The Orchestration Layer What the agent framework provides
person
User Input "Find flights
to Tokyo"
enters
account_tree
Agent Framework Routes, manages
state, calls LLM
calls
psychology
LLM + Tools Reason, act,
observe results
returns
chat
Response Streamed back
to the user
3
arrow_downward LangChain: the foundation library
C LangChain The standard library for LLM applications
edit_note
Prompts ChatPromptTemplate
with variables
pipe |
smart_toy
Models ChatOpenAI,
ChatAnthropic, etc.
pipe |
output
Output Parsers StrOutputParser,
PydanticOutputParser
= chain
link
Chain prompt | model
| parser
4
arrow_downward LangGraph: stateful orchestration on top
D LangGraph Graph-based agent orchestration with state
data_object
State TypedDict holding
messages, data
read
hub
Nodes Functions that
process state
route
call_split
Edges Conditional routing
between nodes
cycle
loop
Cycles Think → Act →
Observe → Decide
5
arrow_downward Chainlit: the chat UI layer
E Chainlit Purpose-built chat interface for agents
chat_bubble
Chat UI Message input,
markdown rendering
shows
stream
Streaming Token-by-token
response display
reveals
visibility
Agent Steps Tool calls, thinking,
intermediate results
6
arrow_downward LangSmith: observability across everything
F LangSmith Tracing, debugging, and evaluation
timeline
Traces Every LLM call,
tool invocation
shows
bug_report
Debug Inputs, outputs,
latency per step
measures
monitoring
Evaluate Quality scores,
cost, token usage
7
arrow_downward How the four layers stack together
G The Full Stack Each layer's responsibility
chat_bubble
Chainlit UI Layer
user sees this
sends to
account_tree
LangGraph Orchestration
state + routing
uses
link
LangChain LLM calls
tools, parsers
traced by
monitoring
LangSmith Observability
across all layers
8
arrow_downward Trace a real request through every layer
H Event Flow Walkthrough "Find me flights to Tokyo" — end to end
8
keyboard
1. User Types Chainlit captures
the message
invoke
play_circle
2. Graph Starts LangGraph reads
state, picks node
LLM
psychology
3. LLM Reasons Decides to call
search_flights tool
tool
flight
4. Tool Executes API call returns
flight results
stream
done_all
5. Response Streamed back
via Chainlit
1
Detail