Day 10 — LangChain | AI Engineering Internship

0–5 min
Real Hook

5–25 min
Core Concept

25–45 min
Live Coding

45–55 min
You Code It

55–60 min
Cheat Sheet

🔥 Real-World Hook (0–5 min)

A raw LLM API call gives you text. But real AI products need: structured JSON output, conversation memory, tool calls, retries on failure, streaming responses.

LangChain is the plumbing that connects an LLM to the rest of your application. Like Django for web — you could build everything from scratch, but why?

📈 LangChain Building Blocks (5–25 min)

💡 The 4 Core Primitives

PromptTemplate — parameterized prompt with {variables}.
LLM / ChatModel — the model call (ChatGroq, ChatOpenAI, etc.).
OutputParser — coerce raw text into Python objects (str, JSON, Pydantic).
Chain (|) — pipe them together: prompt | llm | parser.

💻 Live Demo — Course Q&A Bot (25–45 min)

>from langchain_groq import ChatGroq
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.messages import HumanMessage, SystemMessage
import os

llm = ChatGroq(
    model="llama3-8b-8192",
    api_key=os.getenv("GROQ_API_KEY"),
    temperature=0.3
)

# 1. Simple chain: prompt | llm | parser
prompt = ChatPromptTemplate.from_messages([
    ("system", "You are an AI tutor for Core2CloudX internship students.
         Answer clearly in 3 sentences max. Always give a code example."),
    ("human",  "{question}")
])
parser = StrOutputParser()
chain  = prompt | llm | parser

answer = chain.invoke({"question": "What is gradient descent?"})
print(answer)

# 2. Structured output with Pydantic
from pydantic import BaseModel
from langchain_core.output_parsers import JsonOutputParser

class ConceptCard(BaseModel):
    concept:     str
    one_line:    str
    analogy:     str
    code_example: str

parser2 = JsonOutputParser(pydantic_object=ConceptCard)
prompt2 = ChatPromptTemplate.from_template(
    "Explain the ML concept: {concept}\n{format_instructions}"
)
chain2 = prompt2 | llm | parser2
card = chain2.invoke({
    "concept": "Overfitting",
    "format_instructions": parser2.get_format_instructions()
})
print(card)

📋 You Code It (45–55 min)

✍ Task

Build a ConversationChain that remembers context. Ask it: "My name is Priya", then in the next turn ask "What is my name?" — it should recall. Use ChatMessageHistory from LangChain.

⚡ Day 10 Cheat Sheet

ChatPromptTemplate.from_messages()

Multi-role prompt

prompt | llm | parser

LCEL chain (pipe syntax)

chain.invoke({"var": val})

Run chain with inputs

chain.stream({...})

Stream tokens as they arrive

StrOutputParser()

Raw text output

JsonOutputParser(pydantic_object=M)

Typed JSON output

format_instructions

Tell LLM how to format

ChatMessageHistory()

Conversation memory

← Day 09 Day 11 · Vector DB & RAG →