Overview

The Information Tracker Service is a specialized service designed to extract structured information from unstructured text using LLMs. It provides both schema-based and free-form extraction capabilities, making it highly flexible for various information extraction tasks.

• Schema-Based Extraction: Support for JSON schema validation and structured data extraction
• Free-Form Mode: Flexible extraction without schema constraints
• Context-Aware Processing: Ability to update and extend existing information
• Configurable Output: Control over extraction parameters and model behavior
• Enum Validation: Strict validation for enumerated values
• Field Mapping: Smart mapping between existing and new information

Traditional information extraction often involves regular expressions, rule-based systems, named entity recognition, and manual parsing. The Information Tracker Service improves this by using LLMs for intelligent extraction, supporting flexible schemas, maintaining context awareness, and providing automatic validation.

Endpoints

Process Endpoint

The process endpoint performs information extraction and structured data processing.

API Endpoint

Endpoint Parameters

Request

content (string, required)

The text to extract information from.

context (object, optional)

Optional context and current information:

• updateType (string): Type of information update
• currentInfo (object): Existing information to be updated or extended

config (object, optional)

Schema and model configuration:

• modelId (string): Model to use for extraction (default: "mistral-large-2411")
• temperature (number): Controls extraction precision (default: 0.1, range: 0.0-1.0)
• maxTokens (integer): Maximum tokens for model generation (default: 2000)
• schema (object): JSON schema defining expected data structure

Response

status (string)

Status of the processing ("success" or "error").

result (object)

For successful responses:

• Extracted information according to schema or free-form
• Field structure depends on configuration

metadata (object)

Processing metadata:

• model (string): Model used
• created_at (string): Creation timestamp
• processed_at (string): Processing timestamp
• instance_id (string): Instance identifier
• tracking_id (string): Unique tracking ID
• mode (string): Extraction mode used ("schema" or "free_form")

Examples

Request

Full Request with Context and Schema

Schema-Only Request

Free-Form Extraction

Context-Aware Without Schema

Minimal Configuration

Response

Successful Schema-Based Response

Free-Form Response

Error Response

Schema Features

The service supports various schema features for structured information extraction:

Enum Validation

Array Fields

Nested Objects

Best Practices

Schema Design

• Define clear field names and types
• Use enums for constrained values
• Include field descriptions
• Specify required fields

Context Usage

• Provide relevant current information
• Use appropriate update types
• Include previous state when updating

Configuration Tuning

• Use low temperature for precise extraction
• Set appropriate token limits
• Include descriptive schema fields

Limitations

• Schema must be valid JSON Schema
• Enum values must be exact matches
• Some complex validations not supported