🦜🔗 LangSmith Learning Journey

A structured, hands-on progression through the LangChain ecosystem — from a single LLM call all the way to stateful multi-agent workflows with LangGraph.

---

📚 What This Repo Is

This repository documents my learning path through LangChain, LangGraph, and LangSmith — building progressively more complex AI systems step by step. Each numbered file is a self-contained lesson that builds on the previous one.

The progression follows this arc:

Simple LLM Call → Sequential Chains → RAG (v1→v4) → Agents → LangGraph

---

🗂️ File Structure

langsmith/
├── 1_simple_llm_call.py      # Lesson 1: Basic LLM chain with PromptTemplate
├── 2_sequential_chain.py     # Lesson 2: Multi-step sequential chains (LCEL)
├── 3_rag_v1.py               # Lesson 3a: RAG - naive retrieval baseline
├── 3_rag_v2.py               # Lesson 3b: RAG - improved chunking & embeddings
├── 3_rag_v3.py               # Lesson 3c: RAG - FAISS vector store + retriever
├── 3_rag_v4.py               # Lesson 3d: RAG - conversational memory + history
├── 4_agent.py                # Lesson 4: ReAct agent with DuckDuckGo + tools
├── 5_langgraph.py            # Lesson 5: Stateful graph-based agent (LangGraph)
├── islr.pdf                  # Knowledge base: Introduction to Statistical Learning
├── requirements.txt          # All dependencies (pinned)
└── .gitignore

---

🧭 Lesson Breakdown

1️⃣ 1_simple_llm_call.py — Simple LLM Call

The foundation. Builds a minimal chain using LangChain Expression Language (LCEL):

PromptTemplate → ChatOpenAI → StrOutputParser

Concepts: PromptTemplate, ChatOpenAI, StrOutputParser, pipe operator (|), .invoke()

---

2️⃣ 2_sequential_chain.py — Sequential Chains

Chains multiple LLM steps together so the output of one becomes the input of the next.

Concepts: Multi-step LCEL chains, prompt composition, chaining transforms

---

3️⃣ 3_rag_v1.py → 3_rag_v4.py — Retrieval-Augmented Generation (RAG)

A four-version deep dive into RAG using islr.pdf (Introduction to Statistical Learning) as the knowledge source.

Version	What's new
v1	Naive RAG — load PDF, split, embed, query
v2	Better chunking strategy, improved retrieval quality
v3	FAISS vector store, persistent index, similarity search
v4	Conversational RAG with chat history & memory

Concepts: PyPDFLoader, RecursiveCharacterTextSplitter, OpenAIEmbeddings, FAISS, RetrievalQA, ConversationalRetrievalChain

---

4️⃣ 4_agent.py — ReAct Agent with Tools

An autonomous agent that decides which tools to call based on the query — including live web search.

Tools used:

• 🔍 DuckDuckGoSearchRun — real-time web search
• 📺 YouTubeTranscriptTool — fetch and query YouTube transcripts

Concepts: create_react_agent, tool binding, agent executor, tool schemas

---

5️⃣ 5_langgraph.py — Stateful Agent with LangGraph

The most advanced lesson. Builds a graph-based agent with persistent state, conditional edges, and SQLite checkpointing using LangGraph.

[User Input] → [Agent Node] → (tool needed?) → [Tool Node] → [Agent Node] → [Output]
                                     ↓ no
                                  [END]

Concepts: StateGraph, ToolNode, MessagesState, conditional edges, SqliteSaver checkpointer, streaming

---

⚙️ Setup & Installation

1. Clone the repo

git clone https://github.com/codeantik/langsmith.git
cd langsmith

2. Create a virtual environment

python -m venv venv
source venv/bin/activate        # macOS/Linux
venv\Scripts\activate           # Windows

3. Install dependencies

pip install -r requirements.txt

4. Set up environment variables

Create a .env file in the root:

OPENAI_API_KEY=your_openai_api_key_here
LANGCHAIN_API_KEY=your_langsmith_api_key_here
LANGCHAIN_TRACING_V2=true
LANGCHAIN_PROJECT=langsmith-learning

Get your LangSmith API key at smith.langchain.com

---

▶️ Running the Scripts

# Run any lesson individually
python 1_simple_llm_call.py
python 2_sequential_chain.py
python 3_rag_v1.py
python 4_agent.py
python 5_langgraph.py

---

🔑 Key Dependencies

Package	Purpose
langchain	Core chains, prompts, output parsers
langchain-openai	OpenAI model integration
langchain-community	Community tools (DuckDuckGo, YouTube, etc.)
langgraph	Stateful agent graphs
langsmith	Tracing, evaluation, observability
faiss-cpu	Vector similarity search
pypdf	PDF loading for RAG
openai	Direct OpenAI API client
python-dotenv	Environment variable management
streamlit	(Optional) interactive UI

---

🔭 Observability with LangSmith

All chains and agents are automatically traced in LangSmith when LANGCHAIN_TRACING_V2=true is set. This gives you:

• Full run traces (inputs, outputs, latency)
• Token usage per step
• Error debugging
• Prompt evaluation and comparison

View your traces at → smith.langchain.com

---

🗺️ Learning Roadmap

• Simple LLM calls with LCEL
• Sequential chains
• RAG pipeline (v1 → v4)
• ReAct agents with tools
• Stateful graphs with LangGraph
• Multi-agent systems
• Human-in-the-loop workflows
• LangGraph Studio integration
• Custom evaluators in LangSmith

---

👨‍💻 Author

Ankit Singh — Full Stack Developer & Agentic AI Specialist

---

📄 License

This project is open source and available under the MIT License.