JSON Schema Generator

What inference does well

• Types: string, number, integer, boolean, null, array, object.
• Nullability: a field that is null in one sample and a string in another becomes ["string", "null"].
• Array items: homogeneous arrays produce a single items schema; heterogeneous arrays produce prefixItems.
• Enums: if all observed values are from a small set (configurable, default 10 distinct values), emits an enum.
• Required: with multiple samples, the intersection of keys becomes required; with one sample, all keys are required unless you opt out.
• Formats: strings matching ISO-8601 dates, emails or URIs get an inferred format.

What inference cannot know

• Intent vs example: a sample age: 25 infers type: integer, but cannot know you also accept null. Pass multiple samples that cover edge cases.
• Constraints: minLength, maximum, pattern, you have to add these by hand. Inference does not guess limits from samples.
• Business logic: “exactly one of these three fields must be set” requires oneOf, not inferrable.
• References: the generator emits a flat schema. If you want to factor repeated shapes into $defs, do that after generation.

Example output

From a single sample:

{ "name": "Alice", "age": 30, "tags": ["admin", "user"] }

The inferred schema (draft 2020-12):

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "type": "object",
  "properties": {
    "name": { "type": "string" },
    "age": { "type": "integer" },
    "tags": { "type": "array", "items": { "type": "string" } }
  },
  "required": ["name", "age", "tags"]
}

Common mistakes

• Inferring from one sample. The schema will over-fit, every field becomes required, no null tolerance. Always feed at least 5-10 varied samples.
• Using integer when you meant number. If any sample has a decimal, the inferred type becomes number; if all are whole, it becomes integer. For fields that could be either, include a decimal sample.
• Forgetting optional fields. A field present in 4 of 5 samples but missing from 1 becomes optional, intended. If all 5 samples happen to include it, the schema will mark it required even though it is actually optional in your API.