A comprehensive roadmap and resource collection for learning Generative AI with practical implementation using LangChain. This repository serves as a guided journey from basic concepts to advanced applications, featuring the latest LangChain v1.2.x architecture, agentic AI patterns, and production-ready GenAI systems.
This roadmap has been updated for March 28, 2026 to reflect the latest advancements in Generative AI and LangChain ecosystem.
- ✅ LangChain v1.2.13 – Latest stable release (v1.2.13) published March 19, 2026, with
create_agentimprovements,init_chat_modelLangSmith integration metadata, and middleware runtime export fixes - ✅ LangChain-Core v1.2.22 – langchain-core latest version 1.2.22, released March 24, 2026
- ✅ Latest LLMs – GPT-5.4 (March 2026), Claude Opus 4.6 & Sonnet 4.6 (Feb 2026), Gemini 3.1 Pro (Feb 2026), Llama 4, DeepSeek-V4 comparisons and integration
- ✅ LangSmith Fleet – LangSmith Agent Builder has been renamed to LangSmith Fleet as of March 2026
- ✅ DeepAgents v0.4 – deepagents v0.4 ships pluggable sandbox support, smarter conversation history summarization, and Responses API as the default for OpenAI models
- ✅ Advanced RAG – Multi-agent RAG, multimodal retrieval, and dynamic knowledge updating
- ✅ New Courses & Certifications – LangChain Academy certifications, updated DeepLearning.AI courses
- Overview
- 2026 GenAI Landscape
- GenAI Roadmap
- LangChain Integration
- Top Resources
- Getting Started
- Project Structure
- Examples
- Contributing
- License
This repository provides a structured learning path for developers interested in Generative AI with a focus on practical implementation using LangChain. It contains curated notes, code examples, and implementation guides covering the complete GenAI stack from foundations to production deployment.
| Model | Provider | Context Window | Key Strengths | Best For |
|---|---|---|---|---|
| GPT-5.4 Thinking/Pro | OpenAI | 1,000,000 tokens | State-of-the-art reasoning, computer use, Tool Search, 33% fewer errors vs GPT-5.2 | Enterprise, agentic workflows, complex professional tasks |
| GPT-5.3 Instant | OpenAI | ~200,000 tokens | Fast everyday responses, better web results | Quick-turn tasks, customer support, drafting |
| Claude Opus 4.6 | Anthropic | 1,000,000 tokens | Deepest reasoning, longest autonomous task horizon (14.5h), coding, computer use | Research, long-running agents, code review, cybersecurity |
| Claude Sonnet 4.6 | Anthropic | 1,000,000 tokens (beta) | Near-Opus performance, improved computer use, cost-efficient | Default daily driver, coding, document analysis |
| Gemini 3.1 Pro | 1,000,000 tokens | Reasoning-first, agentic workflows, adaptive thinking | Complex multimodal, agentic coding | |
| Gemini 3 Flash / 3.1 Flash | 1,000,000 tokens | PhD-level reasoning at flash speed, multimodal | Cross-format tasks, real-time apps, cost-sensitive | |
| Llama 4 (Scout/Maverick/Behemoth) | Meta | 10M (Scout) / 1M (Maverick) | Open-weight, MoE architecture, multimodal | Self-hosted, fine-tuning, cost-efficient deployment |
| DeepSeek-V4 | DeepSeek | 128,000+ tokens | Engram memory architecture, strong coding, cost-effective | Budget-conscious deployments, coding, reasoning |
- 🆕 OpenAI released GPT-5.4 on March 5, 2026, billed as "our most capable and efficient frontier model for professional work," with context windows as large as 1 million tokens. GPT-5.4 is 33% less likely to make errors in individual claims compared to GPT-5.2. OpenAI introduced "Tool Search" for the API, allowing models to look up tool definitions as needed rather than consuming tokens for all tools upfront, resulting in faster and cheaper requests.
- 🆕 OpenAI also introduced GPT-5.4 mini and nano on March 17, 2026.
- 🆕 As of February 13, 2026, ChatGPT retired GPT-5 (Instant and Thinking). GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini have also been retired from ChatGPT.
- 🆕 Anthropic released Opus 4.6 on February 5, 2026. As of February 20, 2026, it has the longest task-completion time horizon as estimated by METR, with a 50%-time horizon of 14 hours and 30 minutes.
- 🆕 Anthropic released Sonnet 4.6 on February 17, 2026, priced the same as Sonnet 4.5. Highlights include near-Opus-level performance on coding and document comprehension, dramatically improved computer use capabilities, and better instruction following.
- 🆕 In February 2026, Anthropic introduced Claude Code Security, which reviews codebases to identify vulnerabilities.
- 🆕 Claude now supports computer use in Cowork and Claude Code — users on Pro and Max plans can give Claude access to their computer to open files, run dev tools, point, click, and navigate.
- 🆕 On February 19, 2026, Google released Gemini 3.1 Pro. It is Google's latest reasoning-first model optimized for complex agentic workflows and coding, with adaptive thinking and a 1M token context window.
- 🆕 Gemini 3 Flash is now the default model in the Gemini app, offering next-generation intelligence at lightning speed with PhD-level reasoning comparable to larger models.
- 🆕 DeepSeek V4, slated for mid-February 2026 release, features a novel Engram memory architecture for more efficient data recall and context management.
- 🆕 Meta released Llama 4 on April 5 (expected) — Llama 4 Maverick is a multimodal Mixture-of-Experts model built for textual and visual understanding and efficient deployment.
- 🔮 Anthropic confirmed a new in-development model called "Claude Mythos," described as "the most capable we've built to date," currently being piloted with early customers. A new product tier called "Capybara" will be added above Opus when Claude Mythos launches.
- Agentic AI Revolution – AI agents now autonomously manage workflows, interact with tools, use computers directly, and orchestrate multi-step processes
- Computer Use & Tool Search – Models can now control browsers, desktops, and software UIs; API-level tool search reduces token overhead
- Multimodal by Default – Models seamlessly process text, images, video, audio, and code; native image/video generation built in
- Smaller, Greener Models – Quantization, MoE architectures, and pruning enable efficient edge deployment (GPT-5.4 mini/nano, Gemini Flash Lite)
- AI-Native Enterprises – GenAI embedded in core business operations with autonomous coding agents (Claude Code, Codex)
- Ethics as Engineering – Responsible AI practices built directly into development pipelines; cybersecurity implications of frontier models driving cautious rollouts
-
Machine Learning Basics
- Supervised vs. Unsupervised Learning
- Neural Networks Fundamentals
- Training and Evaluation Metrics
-
NLP Fundamentals
- Text Processing Techniques
- Word Embeddings (Word2Vec, GloVe, FastText)
- Modern Tokenization (BPE, SentencePiece)
-
Deep Learning for NLP
- RNNs, LSTMs, and GRUs
- Attention Mechanisms
- Transformers Architecture (The Foundation of Modern AI)
-
Transformer-Based Models
- GPT Family (GPT-4, GPT-5, GPT-5.2, GPT-5.3 Instant, GPT-5.4 Thinking/Pro)
- Claude Series (Sonnet 4.5/4.6, Opus 4.5/4.6, upcoming Mythos/Capybara)
- Gemini Family (2.5 Pro/Flash, 3 Pro/Flash, 3.1 Pro/Flash)
- Open-Source: Llama 4 (Scout, Maverick, Behemoth), Mistral Medium 3, DeepSeek-V4, Qwen
-
Multimodal Models
- Vision-Language Models (GPT-5.4 Vision, Gemini 3 Vision, Claude computer use)
- Image Generation (DALL-E 3, Stable Diffusion 3, Midjourney v6, Gemini Nano Banana 2)
- Video Generation (Sora, Runway Gen-3, Google Veo 3)
- Audio Models (Whisper, ElevenLabs, Gemini 2.5 Flash TTS)
-
Fine-tuning Strategies
- Transfer Learning
- Prompt Engineering & Optimization
- PEFT (LoRA, QLoRA, Prefix Tuning)
- RLHF & DPO (Direct Preference Optimization)
- Distillation & Quantization
- MetaP Hyperparameter Transfer (Llama 4)
Updated for LangChain v1.2.13 (March 2026)
-
LangChain v1.2.x Basics
create_agentabstraction: the fastest way to build an agent with any model provider, built on LangGraph runtime for reliability- Components and Unified Namespace
- Chains, Agents, and Memory Types
- LangChain Expression Language (LCEL)
- Improved structured output generation integrated directly into the main agent loop, reducing both latency and cost
-
Middleware & Guardrails
- Built-in middleware for human-in-the-loop, summarization, and PII redaction; support for custom middleware to hook into any point in agent execution
- PIIMiddleware (Data Redaction)
- SummarizationMiddleware (Context Management)
- HumanInTheLoopMiddleware (Approval Workflows)
- Content Moderation Middleware
- Model Retry with Exponential Backoff
-
Prompt Engineering with LangChain
- Template Creation & Management
- Few-shot and Zero-shot Learning
- Chain of Thought Prompting
- Dynamic Prompt Construction
-
Advanced LangChain Features
- Document Loading and Intelligent Splitting
- Vector Stores (Chroma, Pinecone, Weaviate, FAISS)
- Embeddings (OpenAI, Cohere, HuggingFace)
- Retrieval Augmented Generation (RAG)
- Tool and API Integration
- MCP Adapters for Multimodal Tools
- Provider integrations: anthropic, aws, azure-ai, deepseek, google-genai, google-vertexai, groq, huggingface, mistralai, ollama, openai, perplexity, together, xai
New Section for 2026
-
LangGraph Fundamentals
- Use LangGraph for advanced needs requiring a combination of deterministic and agentic workflows, heavy customization, and carefully controlled latency
- State Machines for AI Workflows
- Multi-Agent Orchestration
- Conditional Branching & Loops
- Human-in-the-Loop Patterns
-
DeepAgents (v0.4)
- Pluggable sandbox support, smarter conversation history summarization, and Responses API as default for OpenAI models
- Long-running Autonomous Workflows
- Pluggable Storage Backends (S3, Cloud)
- Remote Sandboxes for Security
- Composite Agent Architectures
-
Building Production Agents
- Task Decomposition Patterns
- Tool Selection & Execution (including OpenAI Tool Search)
- Error Recovery & Fallbacks
- Dynamic tools, recovery from hallucinated tool calls, and better streaming error signals
- Agent Memory & State Persistence
Updated RAG Best Practices for 2026
-
RAG Architecture Design
- Chunking Strategies (256-512 tokens optimal)
- Embedding Model Selection
- Hybrid Search (Vector + Keyword)
- Re-ranking for Precision
-
Production RAG Patterns
- Multi-step RAG Pipelines
- Context Condensation
- Source Citation & Traceability
- Dynamic Knowledge Updating
-
Multimodal RAG
- Image & Document Understanding
- Table Extraction & Processing
- Cross-Modal Retrieval
-
Building Conversational Agents
- Chatbots with Memory & Context
- Multi-turn Dialogue Management
- Task-specific Agentic Systems
- Computer-Use Agents (browser automation, desktop control)
-
Content Generation Systems
- Text Summarization
- Creative Writing Assistants
- Code Generation & Review
- Report Generation
-
Information Retrieval & Knowledge Systems
- Enterprise Q&A Systems
- Knowledge Base Construction
- Document Analysis Pipelines
- Semantic Search Engines
-
Model Optimization
- Quantization (INT8, INT4)
- Distillation
- Inference Optimization (vLLM, TensorRT)
-
Deployment Strategies
- API Development (FastAPI, LangServe)
- Containerization (Docker, Kubernetes)
- Serverless Deployment (AWS Lambda, Cloud Functions)
- Edge Deployment
-
Observability with LangSmith
- Insights Agent automatically analyzes your traces to detect usage patterns, common agent behaviors and failure modes
- LangSmith Fleet (formerly Agent Builder) – describe what you want and it figures out the approach, including prompt, tool selection, subagents, and skills
- Pin any experiment as your baseline so every subsequent run is automatically measured against it
- Trace Debugging & Analysis
- Performance Monitoring
- Cost Tracking & Optimization
- A/B Testing with Pairwise Annotation Queues
- LangSmith Fetch CLI for IDE Debugging
LangChain is the easiest way to start building agents and applications powered by LLMs — with under 10 lines of code you can connect to OpenAI, Anthropic, Google, and more. LangChain provides a pre-built agent architecture and model integrations to help you get started quickly.
This repository demonstrates how to leverage LangChain for:
- Building Complex Reasoning Chains with LCEL
- Creating Domain-Specific Chatbots with custom memory
- Implementing Production RAG Systems with hybrid retrieval
- Developing Autonomous Agents with LangGraph
- Connecting LLMs to External Tools and APIs via MCP Adapters
- Deploying Secure, Observable AI Systems with LangSmith Fleet
# create_agent API (LangChain v1.2.13)
from langchain import create_agent
from langchain.middleware import (
PIIMiddleware,
SummarizationMiddleware,
HumanInTheLoopMiddleware
)
agent = create_agent(
model="gpt-5.4", # Updated: GPT-5.4 is the latest frontier model
tools=[search_tool, calculator_tool],
middleware=[
PIIMiddleware(),
SummarizationMiddleware(max_tokens=4000),
HumanInTheLoopMiddleware(require_approval=["sensitive_action"])
]
)- LangChain 0.3 is now in MAINTENANCE mode with support until December 2026, including only security patches and critical bug fixes
- Legacy functionality has been moved to langchain-classic for backwards compatibility
- langchain-community has been released as version 0.4 to reflect a different stability policy where minor releases may include breaking changes
- LangChain Documentation
- LangChain GitHub Repository
- LangChain v1.x Migration Guide
- LangGraph Documentation
- LangSmith Platform
- LangChain Changelog (2026)
- LangChain PyPI – v1.2.13
- LangChain-Core PyPI – v1.2.22
| Course | Platform | Level | Certificate |
|---|---|---|---|
| LangChain for LLM Application Development | DeepLearning.AI | Beginner | ✅ |
| LangChain Academy - Foundation Track | LangChain | Beginner-Advanced | ✅ Official |
| LangChain Academy - DeepAgents Track | LangChain | Advanced | ✅ Official |
| Agentic AI with LangChain and LangGraph | IBM/Coursera | Intermediate | ✅ |
| Developing LLM Applications with LangChain | DataCamp | Intermediate | ✅ |
| LangChain Mastery Course | Udemy | All Levels | ✅ |
| Free LangChain Basics | Great Learning | Beginner | ✅ Free |
- 🎤 Interrupt 2026, the AI Agent Conference — May 13-14
- "Building LLM Powered Applications" by Simon Willison
- "Natural Language Processing with Transformers" by Lewis Tunstall, Leandro von Werra, and Thomas Wolf
- "Generative Deep Learning" (2nd Edition) by David Foster
- "Transformers for Natural Language Processing" by Denis Rothman
- "LangChain in Action" by Harrison Chase (2025)
- "Designing Large Language Model Applications" by O'Reilly (2025)
- LangChain Cookbook
- Building LLM Applications for Production by Chip Huyen
- Prompt Engineering Guide
- RAG Application Development Guide 2026
- The Illustrated Transformer
- LangChain Best Practices (2026)
- Attention Is All You Need - Transformer architecture
- Language Models are Few-Shot Learners - GPT-3 paper
- Training Language Models to Follow Instructions with Human Feedback - InstructGPT/RLHF
- Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks - RAG paper
- Engineering the RAG Stack (2026) - Latest RAG architecture review
- Direct Preference Optimization - DPO paper
- LangSmith - Observability & debugging (now with Fleet & Insights Agent)
- Pinecone - Vector database
- ChromaDB - Open-source embeddings database
- Weaviate - Vector search engine
- vLLM - High-throughput inference
- Claude Code - Agentic coding CLI by Anthropic
- Python 3.10 or higher (recommended: 3.11+; supports up to Python 3.14)
- pip (Python package manager)
- OpenAI API key (or other LLM provider keys)
- Clone this repository:
git clone https://github.com/AdilShamim8/GenAI-Roadmap-with-Notes-Using-LangChain.git
cd GenAI-Roadmap-with-Notes-Using-LangChain- Create and activate a virtual environment:
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate- Install required packages:
pip install -r requirements.txt- Set up environment variables:
# Create a .env file with your API keys
cat > .env << EOF
OPENAI_API_KEY=your_openai_api_key
ANTHROPIC_API_KEY=your_anthropic_api_key
LANGCHAIN_API_KEY=your_langsmith_api_key
LANGCHAIN_TRACING_V2=true
EOFGenAI-Roadmap-with-Notes-Using-LangChain/
├── foundations/ # Basic concepts and foundational knowledge
│ ├── nlp_basics/ # NLP fundamentals
│ ├── transformers/ # Transformer architecture notes
│ └── llm_concepts/ # LLM theory and concepts
├── langchain_basics/ # Introduction to LangChain v1.2.x
│ ├── components/ # Core components of LangChain
│ ├── chains/ # Building and using chains
│ ├── memory/ # Working with different memory types
│ └── middleware/ # Middleware patterns
├── langchain_advanced/ # Advanced LangChain implementations
│ ├── lcel/ # LangChain Expression Language
│ ├── rag/ # Retrieval Augmented Generation
│ ├── agents/ # Building autonomous agents
│ └── tools/ # Tool integration
├── langgraph/ # LangGraph tutorials
│ ├── basics/ # State machines & workflows
│ ├── multi_agent/ # Multi-agent orchestration
│ └── deep_agents/ # Long-running agent patterns (v0.4)
├── projects/ # Complete project implementations
│ ├── chatbot/ # Conversational agent examples
│ ├── document_qa/ # Document Q&A system
│ ├── content_generator/ # Text generation applications
│ └── agentic_assistant/ # Autonomous agent project
├── deployment/ # Deployment guides and examples
│ ├── langserve/ # LangServe API deployment
│ ├── langsmith/ # Observability setup (Fleet + Insights Agent)
│ ├── optimization/ # Model optimization techniques
│ └── monitoring/ # Production monitoring
├── resources/ # Additional learning resources
├── notebooks/ # Jupyter notebooks with examples
├── requirements.txt # Project dependencies
├── .env.example # Example environment variables
└── README.md # This documentation
from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
# Initialize the LLM (GPT-5.4 is the latest as of March 2026)
llm = ChatOpenAI(model="gpt-5.4", temperature=0.7)
# Create a prompt template
prompt = ChatPromptTemplate.from_messages([
("system", "You are a helpful AI assistant."),
("user", "Write a short paragraph about {topic}.")
])
# Create a chain using LCEL
chain = prompt | llm | StrOutputParser()
# Run the chain
result = chain.invoke({"topic": "artificial intelligence in 2026"})
print(result)from langchain_openai import ChatOpenAI, OpenAIEmbeddings
from langchain_community.document_loaders import TextLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_chroma import Chroma
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough
# Load and split documents
loader = TextLoader("path/to/document.txt")
documents = loader.load()
text_splitter = RecursiveCharacterTextSplitter(
chunk_size=512, # Optimal chunk size for 2026
chunk_overlap=50
)
texts = text_splitter.split_documents(documents)
# Create embeddings and vector store
embeddings = OpenAIEmbeddings(model="text-embedding-3-large")
db = Chroma.from_documents(texts, embeddings)
retriever = db.as_retriever(search_kwargs={"k": 5})
# Create RAG chain with source citation
prompt = ChatPromptTemplate.from_template("""
Answer the question based on the following context.
Always cite your sources.
Context: {context}
Question: {question}
Answer:""")
llm = ChatOpenAI(model="gpt-5.4")
rag_chain = (
{"context": retriever, "question": RunnablePassthrough()}
| prompt
| llm
)
# Query with traceability
response = rag_chain.invoke("What are the key points?")
print(response.content)from langgraph.prebuilt import create_react_agent
from langchain_openai import ChatOpenAI
from langchain_core.tools import tool
@tool
def search_web(query: str) -> str:
"""Search the web for information."""
# Implement web search
return f"Search results for: {query}"
@tool
def calculate(expression: str) -> str:
"""Evaluate a mathematical expression."""
return str(eval(expression))
# Create agent with LangGraph
llm = ChatOpenAI(model="gpt-5.4")
agent = create_react_agent(llm, [search_web, calculate])
# Run agent with state management
result = agent.invoke({
"messages": [("user", "What is 25 * 4 and who invented calculus?")]
})
print(result["messages"][-1].content)from langchain import create_agent
from langchain.middleware import (
PIIMiddleware,
SummarizationMiddleware,
HumanInTheLoopMiddleware
)
# Production agent with middleware stack
agent = create_agent(
model="claude-sonnet-4.6", # Or "gpt-5.4", "gemini-3.1-pro"
tools=[search_tool, calculator_tool, file_tool],
middleware=[
PIIMiddleware(),
SummarizationMiddleware(max_tokens=4000),
HumanInTheLoopMiddleware(require_approval=["sensitive_action"])
]
)
result = agent.invoke({"input": "Analyze this quarterly report"})Check the notebooks/ directory for more complete examples and tutorials.
Contributions are welcome! If you'd like to add to this roadmap, improve existing content, or share your implementations:
- Fork the repository
- Create a new branch (
git checkout -b feature/your-feature) - Commit your changes (
git commit -m 'Add some feature') - Push to the branch (
git push origin feature/your-feature) - Open a Pull Request
Please see CONTRIBUTING.md for detailed guidelines.
This project is licensed under the MIT License - see the LICENSE file for details.
- Website: Adil Shamim
- GitHub: Adil Shamim
- Create an issue in this repository for questions or suggestions
⭐ If you find this repository helpful, please consider giving it a star! ⭐
