AI Schema Suggestions

Overview

When ingesting data from new sources, defining field types, merge operations, and stream mappings can be time-consuming. Lytics provides an AI-powered schema suggestion feature that analyzes sample data and recommends appropriate field definitions and mappings.

The schema suggestion system can:

Analyze JSON or CSV sample data from your event streams
Recommend field types (string, number, boolean, date, etc.)
Identify which fields are likely identifiers or PII
Suggest appropriate merge operations for each field
Generate stream mapping expressions

How It Works

Data Analysis: You provide sample records from a stream (or the system pulls them automatically from the event catalog).
AI Processing: The sample data is sent to an LLM (Google Vertex AI or OpenAI) that analyzes field names, values, and patterns.
Suggestion Generation: The AI returns structured suggestions for each field, including type, merge operation, PII classification, and mapping expressions.
Review & Apply: You review the suggestions and apply the ones that fit your schema design.

API Reference

Generate Suggestions from Sample Data

POST /v2/ai/schema/suggest

Analyzes sample records and returns field and mapping suggestions.

Query Parameters

Parameter	Type	Default	Description
`engine`	string	`vertex`	AI engine to use: `vertex` or `openai`
`stream`	string	`default`	Stream name for the suggestions
`table`	string	`user`	Target table name
`format`	string	`json`	Format of the sample data: `json` or `csv`
`prompts`	string[]		Additional custom instructions for the AI
`temperature`	float	`1.05`	Controls randomness of AI output

Request Body

Provide sample records as raw JSON array or CSV data.

[
  {
    "email": "[email protected]",
    "first_name": "Jane",
    "last_name": "Doe",
    "signup_date": "2024-01-15T10:30:00Z",
    "purchase_count": 5,
    "total_spend": 249.99
  }
]

Response

Returns a map of field names to their suggested definitions and mappings:

{
  "email": {
    "fields": [
      {
        "field": "email",
        "is_identifier": true,
        "is_pii": true,
        "shortdesc": "User email address",
        "type": "string",
        "mergeop": "setadd",
        "managed_by": "ai"
      }
    ],
    "mappings": [
      {
        "field": "email",
        "stream": "default",
        "expr": "email",
        "guard_expr": "",
        "managed_by": "ai"
      }
    ]
  },
  "purchase_count": {
    "fields": [
      {
        "field": "purchase_count",
        "is_identifier": false,
        "is_pii": false,
        "shortdesc": "Total number of purchases",
        "type": "int",
        "mergeop": "valuect",
        "managed_by": "ai"
      }
    ],
    "mappings": [
      {
        "field": "purchase_count",
        "stream": "default",
        "expr": "purchase_count",
        "managed_by": "ai"
      }
    ]
  }
}

Retrieve Pre-computed Suggestions for a Stream

GET /v2/ai/stream/{stream}

Returns previously generated field suggestions for a specific stream, if available.

Generate Suggestions for a Specific Field

GET /v2/ai/stream/{stream}/{key}

Generates suggestions for a single field within a stream.

Query Parameters

Parameter	Type	Description
`values`	string[]	Sample values for the field. If omitted, values are fetched from the event catalog.

Response

Returns field and mapping suggestions for the specified key, using the same structure as the full suggestion response.

Suggestion Fields

Each field suggestion includes:

Property	Description
`field`	Recommended field name in the schema
`is_identifier`	Whether the field should be used as an identity key
`is_pii`	Whether the field contains personally identifiable information
`shortdesc`	Human-readable description of the field
`type`	Recommended data type (string, int, float, boolean, date, etc.)
`mergeop`	Recommended merge operation (setadd, valuect, max, min, etc.)
`managed_by`	Set to `ai` indicating the suggestion was AI-generated

Each mapping suggestion includes:

Property	Description
`field`	Target field in the schema
`stream`	Source stream name
`expr`	Mapping expression to extract the value
`guard_expr`	Optional conditional expression for the mapping