# Structured Output – Pydantic

> Extract a validated Pydantic object from raw text without manual parsing.

This repository contains the solution for the exam task *“Structured output (Pydantic)”* from the LangChain course “Generating Structured Outputs”.

---

## 📦 Installation

```bash
# Create and activate a virtual environment (optional but recommended)
python -m venv .venv
source .venv/bin/activate   # Windows: .\.venv\Scripts\activate

# Install dependencies
pip install langchain-core langchain-openai pydantic python-dotenv
```

> **Requirements**  
> * Python 3.10+  
> * LangChain ≥ 1.0.0  

The project uses the OpenAI LLM via `langchain_openai`. If you prefer another provider (e.g., Ollama) replace the import and configuration accordingly.

---

## 📁 Project layout

```
├── solution.py          # Main script – contains models, chain, CLI
└── .env                 # Optional: store your OPENAI_API_KEY here
```

> **.env**  
> Add a line `OPENAI_API_KEY=sk-…` to use the OpenAI API without passing the key on the command line.

---

## 🚀 Running the script

```bash
python solution.py [--text TEXT] [--model MODEL]
```

* `--text TEXT` – raw text to parse.  
  If omitted, the script will prompt you for input or use one of the built‑in examples.
* `--model MODEL` – optional LLM model name (default: `gpt-4o-mini`).  

### Example 1 – Person description

```bash
python solution.py --text "John Doe is a senior software engineer at Acme Corp. He loves Python, Docker and Kubernetes."
```

**Output**

```
Model: PersonInfo
{
  "name": "John Doe",
  "age": null,
  "profession": "senior software engineer at Acme Corp.",
  "skills": ["Python", "Docker", "Kubernetes"]
}
Summary: Parsed as a person profile.
```

### Example 2 – Meeting notes

```bash
python solution.py --text "Meeting on 2024-05-28 with Alice, Bob and Carol. Topics: project roadmap, budget allocation. Decisions: approve Q3 budget, assign tasks to team."
```

**Output**

```
Model: MeetingNotes
{
  "date": "2024-05-28",
  "participants": ["Alice", "Bob", "Carol"],
  "topics": ["project roadmap", "budget allocation"],
  "decisions": ["approve Q3 budget", "assign tasks to team"]
}
Summary: Parsed as meeting notes.
```

---

## 📄 How it works

1. **Pydantic models** – `PersonInfo` and `MeetingNotes`, each field annotated with a description for the LLM.
2. **Prompt template** – instructs the model to output JSON that matches one of the schemas.
3. **Output parser** – `PydanticOutputParser` validates the returned JSON against the chosen schema.
4. **Schema selection** – a simple heuristic (keyword search) decides whether the input describes a person or a meeting, then runs the appropriate chain.

The script prints both the full parsed object (`model_dump()`) and a short human‑readable summary.

---

## 🛠️ Customisation

* Change the LLM model by editing `DEFAULT_MODEL` in `solution.py`.
* Add more heuristics for schema selection.
* Extend the models with additional fields or validation rules.

---

## 📄 License

MIT © 2026 – see [LICENSE](LICENSE) (if present).

---