Nutrient Data Extraction API Delivers Schema‑Aligned JSON with Citations and Confidence
MarkTechPost reports a new API that turns PDFs and Office files into JSON that follows your schema.
Source: MarkTechPostThe Nutrient Data Extraction API announced on November 19, 2026 can read PDFs, scanned images and Office documents and output JSON that mirrors a user‑defined schema, while attaching page‑level citations and confidence scores to every field. According to MarkTechPost, the service also includes a “Studio” environment where teams can prototype and validate the extraction before writing any integration code.
What is the Nutrient Data Extraction API?
The API is a cloud‑based service that focuses on nutritional information embedded in a variety of document formats. It accepts PDFs, scanned files and Microsoft Office documents, runs optical character recognition (OCR) where needed, and extracts the requested data points. The output is a JSON object whose keys and nesting follow the schema supplied by the caller, eliminating the need for downstream transformation.
How does the API generate JSON that matches a custom schema?
Callers provide a schema definition—typically a JSON‑compatible description of the fields they need, such as {"food": "string", "calories": "number", "vitamins": {"A": "number", "C": "number"}}. The service maps the raw text it discovers onto those fields, preserving the hierarchy. If a document contains multiple tables or sections, the API can produce an array of objects that each conform to the same schema. This schema‑first approach saves developers from writing custom parsers for each new document layout.
What are per‑field citations and confidence scores?
Every extracted value is accompanied by metadata that tells the consumer where the information came from. The citation includes the page number, the rectangular region on the page, and a match‑label that identifies the pattern used for extraction. In addition, the API attaches a confidence score—typically a float between 0 and 1—that reflects how certain the model is that the value is correct. This granular provenance lets users audit results, flag low‑confidence entries for manual review, and build more trustworthy data pipelines.
How can developers test the API with Grafy Studio?
Grafy offers a “Studio” environment that mirrors the testing workflow described by the API vendor. Within Grafy Studio, teams can upload sample PDFs or Office files, define the target JSON schema, and run the extraction without writing any code. The interface displays the resulting JSON side‑by‑side with the source document, highlighting the cited regions. This rapid‑feedback loop helps developers fine‑tune their schema and decide whether the confidence thresholds are appropriate before committing to production integration.
What practical use cases does this enable for creators, developers, and researchers?
- Food‑tech startups can ingest nutrition labels from product packaging and feed structured data into recommendation engines.
- Academic researchers conducting meta‑analyses of dietary studies can automatically harvest tables from legacy PDFs, preserving the original page citations for reproducibility.
- Regulatory compliance teams can verify that ingredient lists meet labeling standards by extracting and cross‑checking each field.
- Content creators producing diet‑tracking apps can bulk‑import data from publicly available documents, reducing manual entry time.
How to try the Nutrient Data Extraction API with Grafy tools?
- Upload a document using the Grafy Files app (
/files) to store the PDF or Office file you want to test. - Open Reader (
/reader) to preview the document and confirm that OCR has correctly recognized the text. - Switch to Studio (the testing environment mentioned in the MarkTechPost article) and paste your JSON schema.
- Run the extraction; the resulting JSON will appear with per‑field citations and confidence scores.
- If you need to integrate the API into a larger workflow, use Grafy Code (
/code) to call the endpoint from Python, JavaScript or any language that supports HTTP requests. The code studio can store your API keys securely and let you iterate on error handling based on the confidence values.
What are the limitations and next steps?
The API currently supports only English‑language documents and relies on OCR quality for scanned images. Extremely complex tables or hand‑written notes may produce lower confidence scores, prompting a manual verification step. Future updates are expected to broaden language support and add a batch‑processing mode for high‑volume ingestion.
Source: MarkTechPost – "Nutrient Data Extraction API: JSON Matching Your Schema With Per‑Field Citations and Confidence" (https://www.marktechpost.com/2026/11/18/nutrient-data-extraction-api-citations-confidence/)
Frequently asked questions
Does the API return confidence scores for every extracted field?
Yes. Each field in the output JSON includes a numeric confidence value, allowing users to set thresholds or flag low‑confidence entries for review.
Can I use the API without writing any code?
Grafy’s Studio testing environment lets you upload documents, define a schema, and see the extracted JSON with citations—all through a graphical interface, no coding required.
Is the citation information limited to page numbers?
No. Citations contain the page number, the exact region on the page (coordinates), and a match‑label that indicates the extraction pattern, providing detailed provenance for each value.


