Read forms into typed fields without a template per layout

Superlinked gives your agent one API to turn invoices and scanned forms into typed fields your system can post.

Schema sent, values returned

field	type	returned
form_tracking_number	string	"ACME-12345"
part_number	string	"PW54667"
status_work	string	"See Block 12"
block_13a_approved_design_data_checked	boolean	false
block_14a_other_regulation_checked	boolean	true
signature_date	string	"30 Apr 2008"

Every reply matched the schema on the first call, tick boxes included

View evidence

180 of 223 fields exact across all 8 recorded documents

8 of 8 ticked and empty boxes read correctly

9 of 9 replies valid against the schema, first call

FAA Form 8130-3 release certificate, rebuilt fuel control

Aviation release certificate

form_tracking_number: "ACME-12345"
part_number: "PW54667"
block_13a_approved_design_data_checked: false
block_14a_other_regulation_checked: true
signature_date: "30 Apr 2008"
organization_name: "Acme Airplane Company, 110 Aviation Place, Somewhere, OK (PC62)" expected: "Acme Airplane Company"

Block 14a is ticked for another regulation and block 13a is empty; both arrive as booleans. The organization came back with its address attached.

15 of 16 fields exact 16.2s

FAA Order 8130.21H, figure A-10 · public domain

Certificate of analysis

certified_values[2].element: "Chromium (Cr)" expected: "Chromium"
certified_values[2].symbol: "Cr"
certified_values[2].mass_fraction_percent: 17.803
certified_values[2].expanded_uncertainty_percent: 0.099
certified_values[13].coverage_factor_k: 2.45
srm_number: "SRM 1155a" expected: "1155a"

All 42 certified numbers and all 14 symbols match, and the reference table lower on the page stays out of the answer. The schema asked for element and symbol separately; on all 14 rows the model wrote the printed cell into element and the symbol into symbol, so the name carries the symbol twice. A field description, or one normalization pass after the call, settles it.

56 of 71 fields exact 65.1s

NIST SRM 1155a certificate, page 3 · public domain

Walmart grocery receipt, phone photo taken sideways

Receipt photographed sideways

merchant: "WALMART"
purchase_date: "07/25/21"
total: 7.47 expected: 191.13
subtotal: 6.97 expected: 178.67
items_sold: 21 expected: 77
store_address: "200 N MAIN ST\nBAYARD WY 84713-129" expected: "2000 N WALNUT ST CAMERON MO 64429"

Schema-valid JSON, invented values. The reply carries a different store’s address and phone, 7.47 for a 191.13 total and 21 items for 77. Rotate the page before the call, or check the total against the lines.

4 of 11 fields exact 11.8s

Receipt by Cryptogoth via Wikimedia Commons · CC BY-SA 4.0

180 of 223 fields exact across all 8 recorded documents, in 9 calls. The 5 documents not shown sit in the evidence record, the weakest being DEI electricity bill, Greece, 2015 at 12 of 22.

One call turns a scanned invoice into JSON your ledger can post

View on GitHub

from pathlib import Path
import json
from sie_sdk import SIEClient
client = SIEClient(
    api_key="API keysk-sie-…",
    base_url="https://api.superlinked.com",
)
schema = json.loads("""
schema{
  "type": "object",
  "properties": {
    "invoice_number": { "type": "string" },
    "invoice_date": { "type": "string" },
    "customer_id": { "type": "string" },
    "amount_due": { "type": "number" },
    "currency": { "type": "string" },
    "due_date": { "type": "string" }
  },
  "required": ["invoice_number", "invoice_date", "customer_id", "amount_due", "currency", "due_date"]
}
""")
image = {"data": Path(filePNGwolters-kluwer-invoice.pngbrowse).read_bytes(), "format": "png"}
result = client.generate(
    "modelQwen/Qwen3.8-27B-FP8",
    "Extract the document fields described by the JSON schema. Return only the JSON object.",
    max_new_tokens=512,
    images=[image],
    grammar={"json_schema": schema, "strict": True},
)
print(json.loads(result["text"]))

import { readFile } from 'node:fs/promises';
import { SIEClient } from '@superlinked/sie-sdk';

const client = new SIEClient('https://api.superlinked.com', {
  apiKey: 'sk-sie-…',
});

const schema = {
  "type": "object",
  "properties": {
    "invoice_number": { "type": "string" },
    "invoice_date": { "type": "string" },
    "customer_id": { "type": "string" },
    "amount_due": { "type": "number" },
    "currency": { "type": "string" },
    "due_date": { "type": "string" }
  },
  "required": ["invoice_number", "invoice_date", "customer_id", "amount_due", "currency", "due_date"]
};
const image = await readFile("wolters-kluwer-invoice.png");
const result = await client.generate(
  'Qwen/Qwen3.8-27B-FP8',
  "Extract the document fields described by the JSON schema. Return only the JSON object.",
  {
    maxNewTokens: 512,
    images: [image],
    grammar: { json_schema: schema, strict: true },
  },
);
console.log(JSON.parse(result.text));

document_image=$(base64 < 'wolters-kluwer-invoice.png' | tr -d '\n')
curl https://api.superlinked.com/v1/generate/Qwen__Qwen3.8-27B-FP8 \
  -H "Authorization: Bearer sk-sie-…" \
  -H "Content-Type: application/json" \
  -d "{\"prompt\":\"Extract the document fields described by the JSON schema. Return only the JSON object.\",\"max_new_tokens\":512,\"images\":[{\"data\":\"$document_image\",\"format\":\"png\"}],\"grammar\":{\"json_schema\":{\"type\":\"object\",\"properties\":{\"invoice_number\":{\"type\":\"string\"},\"invoice_date\":{\"type\":\"string\"},\"customer_id\":{\"type\":\"string\"},\"amount_due\":{\"type\":\"number\"},\"currency\":{\"type\":\"string\"},\"due_date\":{\"type\":\"string\"}},\"required\":[\"invoice_number\",\"invoice_date\",\"customer_id\",\"amount_due\",\"currency\",\"due_date\"]},\"strict\":true}}"

Build the "Doc field extraction" capability into my app using the Superlinked Inference Engine (SIE).

Context
- SIE is an OpenAI-style inference API. Python SDK: `from sie_sdk import SIEClient`; TypeScript: `@superlinked/sie-sdk`.
- Base URL: https://api.superlinked.com (or my regional endpoint). Auth: Bearer key from env `SIE_API_KEY` (never hard-code it).
- Model: Qwen/Qwen3.8-27B-FP8 (SIE primitive: /generate). Keep the model id configurable.

Task
- Input: an uploaded document plus the JSON schema for its fields.
- Behaviour: parse the document, then return fields matching the JSON schema
- Call the selected SIE primitive once per request and map the response into your domain type.

Deliverables
- A typed client wrapper, an application-level function for this task, error handling for timeouts/empty input, and unit tests with a stubbed client.
- Wire it into my existing stack (ask me which framework if unclear) and add a short usage example.

Run

Output

{
  "amount_due": 3900,
  "currency": "USD",
  "customer_id": "UCSFD00005",
  "due_date": "11 Sep 2016",
  "invoice_date": "8/12/2016",
  "invoice_number": "000277511"
}

Parses the document before extracting typed fields

PRICE

$ / 1M input tokens $ / 1M output tokens $ / 1k pages ↓

$0.3 Google Gemini 3.5 Flash-Lite

SIE Qwen3.5 4B $0.72

OpenAI GPT-5.4 mini $0.75

$0.72 SIE Qwen3.5 4B

$2.5 Google Gemini 3.5 Flash-Lite

OpenAI GPT-5.4 mini $4.5

Google Document AI $30

AWS Textract Forms $50

QUALITY

Field accuracy ↑

SIE Qwen3.5 4B 0.89

0.87

0.86

0.85

0.84

LATENCY

p50 s ↓

SIE Qwen3.5 4B 0.9s

1.0s

1.1s

1.2s

Deploy your way

Managed Cloud

Full compute toolkit for your agents with zero ops.

No idle GPUs, pay for what you use
Fits your stack: SDK, API, CLI, MCP
Zero lock-in, self-host the same stack
SOC 2 Type 2, US or EU data residency

$50 Example credit balance

1.3B tokens embedded

120k pages parsed

27M tokens generated (1:1 in:out)

Get started

no credit card required

Self-host with K8s

Easy & scalable deployment in your own cloud.

Terraform to your cloud in minutes
Apache-2.0, same engine as Cloud
Scales to zero, no bill between jobs
Per-tenant pools, no noisy neighbors

Agent prompt

Deploy SIE to our AWS account with the superlinked/sie/aws Terraform module. Docs: superlinked.com/docs/deploymentDeploy SIE to our GCP project with the superlinked/sie/google Terraform module. Docs: superlinked.com/docs/deploymentDeploy SIE to our Azure AKS cluster via helm install. Requirements: superlinked.com/docs/deployment

Deploy guide

Run locally

Run the same models on your own machine.

Runs on NVIDIA GPU or Apple Silicon
One command, no Docker or cluster
All 100+ Cloud models, fully offline
Same SDK and IDs, no code changes

pip install "sie-server[local]" && sie-server servepip install "sie-server[local]" && sie-server serve --device cuda

Quickstart