Output Validation — Animate | AI Wisdom

← Back to topic
🎞 ANIMATE · Mode 3 of 5 · +100 XP— LLM Output Validation
📋Schema
⚙️JSON Mode
🔄Parse
✅Validate
🎯 Next: Test →
Step 1 / 4
Speed
Define Output Schema
ProductReview Pydantic model defines the exact structure LLM must return💡 Analogy
Like designing a form with specific fields — the LLM must fill in each field correctly.
Define a Pydantic model for the expected LLM output. This acts as a contract — any output that doesn't match this shape is rejected.
Console Output
> ProductReview schema:
> sentiment: str
> score: int
> summary: str
> key_points: list[str]
●output_validation.py
Define Output SchemaL1–10
1from pydantic import BaseModel, field_validator
2from openai import OpenAI
3import json
4 
5class ProductReview(BaseModel):
6    sentiment: str
7    score: int
8    summary: str
9    key_points: list[str]
10 
11    @field_validator('sentiment')
12    @classmethod
13    def validate_sentiment(cls, v: str) -> str:
14        allowed = {'positive', 'negative', 'neutral'}
15        if v.lower() not in allowed:
16            raise ValueError(f"sentiment must be one of {allowed}")
17        return v.lower()
18 
19    @field_validator('score')
20    @classmethod
21    def validate_score(cls, v: int) -> int:
22        if not 1 <= v <= 10:
23            raise ValueError("score must be 1-10")
24        return v
25 
26def extract_review(review_text: str) -> ProductReview:
27    client = OpenAI()
28    response = client.chat.completions.create(
29        model="gpt-4o-mini",
30        messages=[{
31            "role": "user",
32            "content": f"Analyse this review and return JSON with sentiment (positive/negative/neutral), score (1-10), summary (str), key_points (list[str]).\n\nReview: {review_text}"
33        }],
34        response_format={"type": "json_object"},
35    )
36    raw = json.loads(response.choices[0].message.content)
37    return ProductReview(**raw)   # Pydantic validates the structure
38 
39review = extract_review("Great product! Battery lasts 3 days. A bit expensive but worth it.")
40print(review.sentiment)   # "positive"
41print(review.score)       # 8
41 lines
Timeline — 4 Steps
Define Output Schema
ProductReview Pydantic model defines the exact structure LLM must return💡 Analogy
Like designing a form with specific fields — the LLM must fill in each field correctly.
Define a Pydantic model for the expected LLM output. This acts as a contract — any output that doesn't match this shape is rejected.
Console Output
> ProductReview schema:
> sentiment: str
> score: int
> summary: str
> key_points: list[str]

Was this page helpful?

Discussion

ProductReview Pydantic model defines the exact structure LLM must return

💡 Analogy

Like designing a form with specific fields — the LLM must fill in each field correctly.

Define a Pydantic model for the expected LLM output. This acts as a contract — any output that doesn't match this shape is rejected.

Console Output

> ProductReview schema:
> sentiment: str
> score: int
> summary: str
> key_points: list[str]

ProductReview Pydantic model defines the exact structure LLM must return

💡 Analogy

Like designing a form with specific fields — the LLM must fill in each field correctly.

Define a Pydantic model for the expected LLM output. This acts as a contract — any output that doesn't match this shape is rejected.

Console Output

> ProductReview schema:
> sentiment: str
> score: int
> summary: str
> key_points: list[str]