# LangChain-WS

**Repository Path**: ailearning1/lang-chain-ws

## Basic Information

- **Project Name**: LangChain-WS
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-12-25
- **Last Updated**: 2025-12-25

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# LangChain Workshop - AI Engineering Lifecycle

Enterprise workshop teaching the complete AI engineering lifecycle using LangChain, LangGraph, and LangSmith—centered around building a customer support agent for TechHub (fictional e-commerce store).

## What You'll Build

A customer support agent system featuring:
- **Multi-agent architecture** with specialized Database and Documents agents coordinated by a Supervisor
- **Human-in-the-loop (HITL)** customer verification with LangGraph primitives
- **Evaluation-driven development** using offline evaluation to identify and fix bottlenecks
- **Production deployment** to LangSmith with online evaluation and data flywheels

## Quick Setup

This workshop uses [uv](https://docs.astral.sh/uv/) - a fast Python package installer and resolver.

```bash
# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install dependencies
uv sync

# Configure environment
cp .env.example .env
# Edit .env with your API keys (see labs/README.md for details)

# Launch Jupyter
uv run jupyter lab
```

For detailed setup instructions, see [labs/README.md](labs/README.md).

## Workshop Structure

### Labs

| Lab | Topic | Description |
|-----|-------|-------------|
| **Lab 1** | Agent Foundations | Tool calling, single agents, database & document tools |
| **Lab 2** | Multi-Agent Systems | Supervisor pattern, agent coordination, state management |
| **Lab 3** | Human-in-the-Loop | Customer verification, interrupts, state persistence |
| **Lab 4** | Deploy & Evaluate | LangSmith deployments, SDK, online evaluation, data flywheel |

### Slides

Workshop slides are available in the `slides/` directory:
```bash
cd slides
pnpm install
pnpm dev  # Visit http://localhost:3030
```

## Repository Structure

```
langchain-workshop-porsche/
├── workshop/
│   ├── labs/                    # Interactive Jupyter notebooks
│   │   ├── lab_1.ipynb          # Agent Foundations
│   │   ├── lab_2.ipynb          # Multi-Agent Systems
│   │   ├── lab_3.ipynb          # Human-in-the-Loop
│   │   ├── lab_4.ipynb          # Deploy & Evaluate
│   │   └── *_solution.ipynb     # Solution notebooks
│   │
│   ├── agents/                  # Reusable agent factory functions
│   │   ├── db_agent.py          # Database queries (rigid tools)
│   │   ├── sql_agent.py         # Flexible SQL generation
│   │   ├── docs_agent.py        # RAG for product docs & policies
│   │   ├── supervisor_agent.py  # Multi-agent coordinator
│   │   └── supervisor_hitl_agent.py  # Verification + routing system
│   │
│   ├── tools/                   # Database & document search tools
│   │   ├── database.py          # DB tools (orders, products, SQL)
│   │   └── documents.py         # RAG tools (products, policies)
│   │
│   ├── evaluators/              # Evaluation metrics
│   │   └── evaluators.py        # Correctness & tool call counters
│   │
│   ├── deployments/             # Production-ready graph configurations
│   │   └── *_graph.py           # LangSmith deployment exports
│   │
│   ├── data/                    # Dataset & generation scripts
│   │   ├── structured/          # SQLite DB + JSON files
│   │   ├── documents/           # Markdown docs for RAG
│   │   └── vector_stores/       # Pre-built vectorstore
│   │
│   ├── config.py                # Workshop configuration
│   └── langgraph.json           # LangGraph deployment config
│
└── slides/                      # Slidev presentation
    ├── slides.md                # Presentation source
    └── slides.pdf               # Exported PDF
```

## Configuration

The workshop uses SiliconFlow API (OpenAI-compatible) with DeepSeek models by default:

| Setting | Default Value |
|---------|---------------|
| Model | `openai:deepseek-ai/DeepSeek-V3.2` |
| Embedding | `openai:Qwen/Qwen3-Embedding-0.6B` |
| API Base | `https://api.siliconflow.cn/v1` |

Configure via environment variables in `.env`:
- `WORKSHOP_MODEL` - LLM for agents
- `WORKSHOP_EMBEDDING_MODEL` - Embedding model for RAG
- `OPENAI_API_KEY` - SiliconFlow API key
- `LANGSMITH_API_KEY` - LangSmith API key

## Dataset Overview

The **TechHub dataset** includes:
- **50 customers** across consumer, corporate, and home office segments
- **25 products** (laptops, monitors, keyboards, audio, accessories)
- **250 orders** spanning 2 years with realistic patterns
- **439 order items** with product affinity patterns
- **SQLite database** with full schema
- **30 documents** (25 product specs + 5 policies) for RAG

## Key Concepts Covered

- **Agent Development**: Tool calling, multi-agent systems, supervisor pattern, HITL with interrupts
- **Evaluation & Testing**: Offline evaluation, LLM-as-judge, trace metrics, eval-driven development
- **Deployment & Production**: LangSmith deployments, online evaluation, annotation queues, SDK integration
- **Best Practices**: Factory functions, state management, dynamic prompts, structured outputs, streaming

## Prerequisites

### Required Knowledge

Free courses from [LangChain Academy](https://academy.langchain.com):
- [LangChain Essentials - Python](https://academy.langchain.com/courses/langchain-essentials-python) (30 min)
- [LangGraph Essentials - Python](https://academy.langchain.com/courses/langgraph-essentials-python) (1 hour)
- [LangSmith Essentials](https://academy.langchain.com/courses/quickstart-langsmith-essentials) (30 min)

### Technical Requirements

- Python 3.10+
- API Keys: LangSmith + SiliconFlow (or other model provider)
- Tools: Git, uv

## Resources

- [LangChain Python Docs](https://python.langchain.com)
- [LangGraph Python Docs](https://langchain-ai.github.io/langgraph)
- [LangSmith Platform](https://smith.langchain.com)
- [LangChain Academy](https://academy.langchain.com)