Packaging for the Cloud: Clean Projects

A professional LangChain app is not a single file. It is a Package. If you just send your app.py to a friend, it will fail because they don't have your 15 libraries or your .env file. Proper packaging ensures that your code runs identically on your laptop and in the cloud.

1. The Standard Project Structure

my-ai-app/
├── main.py          # FastAPI entry point
├── chain_logic.py   # Your LangChain definitions
├── requirements.txt # All your pip libraries
├── .env             # Secrets (NOT pushed to GitHub!)
└── data/            # Local data/vector storage

---

2. Managing Dependencies

LangChain updates fast. If you don't "Pin" your versions, your code might break next month.

Bad requirements.txt:

langchain
langchain-openai

Good requirements.txt:

langchain==0.3.0
langchain-openai==0.2.1
python-dotenv
fastapi
uvicorn

---

3. Visualizing the Build Artifact

graph LR
    Src[Source Code] --> Pack[Packaging Tool]
    Deps[Dependencies] --> Pack
    Pack --> Artifact[Distro / Image]
    Artifact --> Production[Cloud Server]

---

4. Handling Environments

You should never hardcode your path to the vector store index. Use environment variables!

• Instead of: db = FAISS.load_local("/Users/sudeep/apps/index")
• Use: db = FAISS.load_local(os.getenv("INDEX_PATH")) This allows you to move the folder on the server without changing the code.

---

5. Engineering Tip: pip freeze

When you have your app working perfectly, run: pip freeze > requirements.txt This automatically captures every single library and version currently in your virtual environment.

---

Key Takeaways

• Version Pinning is mandatory for AI library stability.
• Keep Logic (LangChain) separate from Transportation (FastAPI).
• Use Environment Variables for secrets and file paths.
• Clean project structure is the prerequisite for Dockerization.