LangChain Examples

Practical, minimal examples for building with LangChain and friends (LangGraph, FAISS, Ollama, etc.). Start locally, then adapt to your stack.

Contents

• Main Applications of LangChain
• RAG Demo (Ollama + FAISS)
• Workflows Demo (Map-Reduce, LCEL)
• Troubleshooting

Main Applications

• RAG (Retrieval-Augmented Generation): Answer questions over your docs, wikis, tickets, and codebases using chunking, embeddings, retrievers, re-ranking, and citations.
• Multi-step Workflows: Summarize, extract, translate, and classify at scale using deterministic chains and map-reduce patterns.
• Tool-Using Agents: Safely call APIs, databases, search, and internal tools with plan→act loops (often built with LangGraph for reliability).
• Structured Extraction: Produce typed JSON or fill schemas from semi-structured text via output parsers and validation.
• Conversational AI with Memory: Build chat experiences that remember context and can take actions through function/tool calling.
• Code & Data Assistants: Repo Q&A, refactoring helpers, SQL generation over warehouses/lakes, and “chat with your data.”

RAG Demo (Ollama + FAISS)

A minimal Retrieval-Augmented Generation example lives in RAG/ and lets you ask questions over local .md/.txt files using a Gemma model served by Ollama.

Prerequisites

• Install Ollama and pull models:
  • ollama pull gemma3:1b
  • ollama pull nomic-embed-text
• Python 3.10+

Quickstart

• Create a virtual environment and install deps:
  • python -m venv .venv && source .venv/bin/activate
  • pip install -r RAG/requirements.txt
• Ingest sample docs and build the FAISS index:
  • python RAG/ingest.py
• Ask questions:
  • python RAG/query.py "What are the main applications of LangChain?"
  • or interactive: python RAG/app.py (streams tokens)

Configuration

• Models: LLM_MODEL (default gemma3:1b), EMBED_MODEL (default nomic-embed-text).
• Paths: INDEX_DIR, DATA_DIR (default to subfolders of RAG/).
• Ollama URL: OLLAMA_BASE_URL if not http://127.0.0.1:11434.

Project Structure

RAG/
  README.md           # RAG-specific docs
  requirements.txt    # LangChain, FAISS, dotenv
  ingest.py           # Build local FAISS index from data/
  query.py            # Query the index with Gemma via Ollama
  app.py              # Simple interactive CLI
  data/               # Sample .md/.txt files
  index/              # Generated FAISS index (gitignored)

Workflows Demo (Map-Reduce, LCEL)

Deterministic multi-step pipelines for summarization, structured extraction, translation, and classification using a local Gemma model via Ollama.

Prerequisites

• ollama pull gemma3:1b
• Python 3.10+

Quickstart (async)

• Create venv and install deps:
  • python -m venv .venv && source .venv/bin/activate
  • pip install -r Workflows/requirements.txt
• Try examples with the included sample:
  • python Workflows/summarize.py Workflows/data/multistep_sample.txt --concurrency 4
  • python Workflows/extract.py Workflows/data/multistep_sample.txt --concurrency 4
  • python Workflows/translate.py Workflows/data/multistep_sample.txt --lang es --concurrency 4
  • python Workflows/classify.py Workflows/data/multistep_sample.txt --labels tutorial reference tips --concurrency 4

Configuration

• LLM_MODEL (default gemma3:1b), OLLAMA_BASE_URL for non-default hosts.
• Scripts support --concurrency to control async map parallelism.
• Adjust chunking via --chunk_size/--chunk_overlap where available.

Troubleshooting

• Ollama not reachable: ensure the daemon is running; set OLLAMA_BASE_URL.
• No docs indexed: add .md/.txt files to RAG/data/ and rerun python RAG/ingest.py.
• Import errors: verify you’re in the venv and ran pip install -r RAG/requirements.txt.

References

• LangChain Docs: https://python.langchain.com/
• LangGraph: https://langgraph.readthedocs.io/
• Ollama: https://ollama.ai/