Structured Output Demo (LangChain + Pydantic + TypedDict + Google Gemini)

This repository demonstrates multiple ways to enforce structured outputs from LLMs (LangChain + Google Gemini or HuggingFace) using Python type‑safety tools (Pydantic models, JSON Schema, and TypedDict), plus a simple Pydantic data‑validation example.

---

📁 Repository Structure

File	Description
pydantic_demo.py	Shows basic Pydantic usage: defines a Student model, validates input data, and serializes to JSON.
json_schema.json	A standalone JSON Schema defining a minimal “student” object (title, type, required fields).
structured_output_with_gemini.py	Uses LangChain + HuggingFace endpoint to generate a review summary with a Pydantic schema for structured output.
structured_output_with_json_schema.py	Uses LangChain + Google Gemini with a raw JSON Schema to enforce output structure.
typedict_test.py	Demonstrates Python’s TypedDict for static‑type validated dictionaries.
with_structured_output_pydantic.py	Uses LangChain + Google Gemini with a Pydantic model to produce strongly typed review output.
with_structured_output_typedict.py	Uses LangChain + Google Gemini with a TypedDict schema for structured output and prints a specific field.

---

🚀 Quick Start

1. Clone this repo

   git clone https://github.com/<your-username>/structuredoutput.git
   cd structuredoutput

2. Create & activate a virtual environment

   python3 -m venv venv
   source venv/bin/activate   # macOS/Linux
   venv\Scripts\activate      # Windows

3. Install dependencies

   pip install -r requirements.txt

4. Add your API key to a .env file

   GOOGLE_API_KEY=your_gemini_api_key_here
   

5. Run any example script

   python pydantic_demo.py
   python structured_output_with_gemini.py
   python with_structured_output_typedict.py

---

🔧 Tech Stack

• LangChain — orchestration
• Google Gemini (PaLM) via langchain-google-genai
• Pydantic & TypedDict — structured output schemas
• HuggingFaceEndpoint — alternative LLM backend
• python-dotenv — environment management

---

📄 License

MIT © 2025

📫 Contact

Questions or feedback? Email [email protected].