Extract → Split → Schema

Overview

The Extract → Split → Schema pipeline is the most powerful processing mode in Pulse. After extracting the document, it splits the pages into topic-based sections and then applies a different schema to each section. This is ideal for long, multi-section documents where different parts contain different kinds of data.

When to Use

Annual reports — Financials, Leadership, and Outlook each have different data to extract
Multi-section contracts — different clause types (indemnification, IP rights, payment terms) need different schemas
Research papers — Abstract, Methodology, Results, and Conclusion each have distinct structure
Insurance documents — policy details, claims history, and coverage schedules are all different
Regulatory filings — mixed sections like company overview, financial statements, risk factors

If your entire document uses one schema, use Extract → Schema instead — it’s simpler and faster.

How to Use in the Playground

What You Get Back

Everything from Extract, plus:

Field	Description
`split_output.splits`	Page assignments — `{ "Financials": [1,2,3], "Leadership": [4,5] }`
`split_id`	Saved split result ID
`results`	Per-topic structured output with `values` and `citations` for each topic
`schema_id`	Saved schema result ID

API Usage

Python
TypeScript
curl

from pulse import Pulse

client = Pulse(api_key="YOUR_API_KEY")

# Step 1: Extract
extract_result = client.extract(
    file=open("annual_report.pdf", "rb"),
    async_=True,
    storage={"enabled": True}
)
extraction_id = extract_result.extraction_id

# Step 2: Split
split_result = client.split(
    extraction_id=extraction_id,
    split_config={
        "split_input": [
            {"name": "Financials", "description": "Revenue, expenses, and profit data"},
            {"name": "Leadership", "description": "Executive team and board members"},
            {"name": "Outlook", "description": "Future plans and projections"}
        ]
    }
)
split_id = split_result.split_id
print(f"Split: {split_result.split_output}")

# Step 3: Schema per topic
schema_result = client.schema(
    split_id=split_id,
    split_schema_config={
        "Financials": {
            "schema": {
                "type": "object",
                "properties": {
                    "total_revenue": {"type": "number"},
                    "net_income": {"type": "number"}
                }
            },
            "schema_prompt": "Extract financial metrics"
        },
        "Leadership": {
            "schema": {
                "type": "object",
                "properties": {
                    "ceo": {"type": "string"},
                    "board_members": {"type": "array", "items": {"type": "string"}}
                }
            },
            "schema_prompt": "Extract leadership information"
        },
        "Outlook": {
            "schema": {
                "type": "object",
                "properties": {
                    "guidance": {"type": "string"},
                    "growth_target": {"type": "number"}
                }
            },
            "schema_prompt": "Extract forward-looking statements"
        }
    }
)

for topic, data in schema_result.results.items():
    print(f"{topic}: {data}")

import { PulseClient } from "pulse-ts-sdk";
import fs from "fs";

const client = new PulseClient({
    apiKey: "YOUR_API_KEY"
});

// Step 1: Extract (async)
const extractResult = await client.extract({
    file: fs.createReadStream("annual_report.pdf"),
    async: true,
    storage: { enabled: true }
});
const extractionId = extractResult.extraction_id;

// Step 2: Split
const splitResult = await client.split({
    extraction_id: extractionId,
    split_config: {
        split_input: [
            { name: "Financials", description: "Revenue, expenses, and profit data" },
            { name: "Leadership", description: "Executive team and board members" },
            { name: "Outlook", description: "Future plans and projections" }
        ]
    }
});
const splitId = splitResult.split_id;

// Step 3: Schema per topic
const schemaResult = await client.schema({
    split_id: splitId,
    split_schema_config: {
        Financials: {
            schema: { type: "object", properties: { total_revenue: { type: "number" } } },
            schema_prompt: "Extract financial metrics"
        },
        Leadership: {
            schema: { type: "object", properties: { ceo: { type: "string" } } },
            schema_prompt: "Extract leadership information"
        },
        Outlook: {
            schema: { type: "object", properties: { guidance: { type: "string" } } },
            schema_prompt: "Extract forward-looking statements"
        }
    }
});

console.log(schemaResult.results);

# Step 1: Extract
curl -X POST https://api.runpulse.com/extract \
  -H "x-api-key: YOUR_API_KEY" \
  -F "file=@annual_report.pdf" \
  -F "async=true" \
  -F 'storage={"enabled": true}'
# Save extraction_id from response

# Step 2: Split
curl -X POST https://api.runpulse.com/split \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "extraction_id": "EXTRACTION_ID",
    "split_config": {
      "split_input": [
        {"name": "Financials", "description": "Revenue, expenses, and profit data"},
        {"name": "Leadership", "description": "Executive team and board members"},
        {"name": "Outlook", "description": "Future plans and projections"}
      ]
    }
  }'
# Save split_id from response

# Step 3: Schema per topic
curl -X POST https://api.runpulse.com/schema \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "split_id": "SPLIT_ID",
    "split_schema_config": {
      "Financials": {
        "schema": {"type": "object", "properties": {"total_revenue": {"type": "number"}}},
        "schema_prompt": "Extract financial metrics"
      },
      "Leadership": {
        "schema": {"type": "object", "properties": {"ceo": {"type": "string"}}},
        "schema_prompt": "Extract leadership information"
      }
    }
  }'

Skipping Schema (Extract → Split Only)

You don’t have to add schema after splitting. If you just want to know which pages belong to which topic — without structured extraction — you can stop after the split step. This is useful for document triage or routing.

# Extract → Split only (no schema)
split_result = client.split(
    extraction_id=extraction_id,
    split_config={
        "split_input": [
            {"name": "Relevant", "description": "Pages related to our query"},
            {"name": "Not Relevant", "description": "Background or boilerplate pages"}
        ]
    }
)

print(split_result.split_output)
# {"splits": {"Relevant": [1,3,5], "Not Relevant": [2,4,6]}}

Split API Reference

Full API documentation for the /split endpoint

Schema API Reference

Full API documentation for the /schema endpoint (single and split mode)

Platform Reference

Pipelines

Forms

Production Workflow

Library & Exports

Extract → Split → Schema

Overview

When to Use

How to Use in the Playground

What You Get Back

API Usage

Skipping Schema (Extract → Split Only)

Split API Reference

Schema API Reference

​Overview

​When to Use

​How to Use in the Playground

​What You Get Back

​API Usage

​Skipping Schema (Extract → Split Only)

​Related

Split API Reference

Schema API Reference

Overview

When to Use

How to Use in the Playground

What You Get Back

API Usage

Skipping Schema (Extract → Split Only)

Related