Extract data from specifications and datasheets

A product specification, the datasheet a manufacturer publishes for a component, is a dense grid of parameters, values, units, and tolerances, and it is the document an engineer reaches for when deciding whether a part fits a design. A motor-drive datasheet states an input voltage of 230V plus or minus 10%, an operating temperature range of -10 to 55 C, an IP65 environmental rating, a power draw in watts, physical dimensions and weight, the connector and communication interface, the compliance standards it meets, and a warranty in months. Design engineers checking fit, procurement teams matching a part number against a bill of materials, and quality staff confirming a component meets ISO 9001 and ISO 14001 requirements all read the same sheet for different numbers, and every one of those numbers carries a unit and often a tolerance that changes what the value means. The trouble is that no two manufacturers lay a datasheet out the same way, and the parameters that matter are scattered across tables, footnotes, and revision blocks. One vendor tabulates electrical, mechanical, and environmental specs separately, another folds them into a single block, and a revision letter on the header quietly supersedes last year numbers. A part carries a manufacturer, a product code or SKU, and a revision number that together identify exactly which version of the spec is in hand. Feed a datasheet from Bosch Rexroth revised 2026-04-30 for the IndraDrive Cs series to Talonic and the parameters come back as rows of name, value, unit, and tolerance, the operating conditions as a min and max per condition, the physical dimensions and materials as their own tables, and the compliance certifications with their issuing bodies and dates, so a spec that was locked in a PDF loads into an engineering database as structured JSON or CSV.

What gets extracted from specifications and datasheets

How extraction works for specifications and datasheets

Datasheets arrive as manufacturer PDFs, distributor exports, and scanned catalog pages, and the same parameter hides under a different label and a different table on each. Talonic classifies the datasheet and maps it to the specification model in the Field Registry, which returns the identity fields (manufacturer, product code, revision) as a header and the technical content as typed tables: parameters with value, unit, and tolerance, operating conditions with a min and max, physical dimensions, materials, performance ratings, and compliance certifications. Values keep their units so a torque in newton-metres is never confused with one in pound-feet, tolerances such as plus or minus 5% are captured alongside the nominal value, and standards such as ISO 9001, ISO 14001, or an IP-rated enclosure are read as stated. Every value returns with a confidence score and a pixel-region pointer conforming to DIN SPEC 91491, so an engineer can check a rated voltage or an operating range against the source datasheet before the part is designed in.

Sample extraction

A servo-drive datasheet with electrical, environmental, and compliance tables

{
  "document_date": "2026-04-30",
  "product_name": "IndraDrive Cs servo drive",
  "product_code": "HCS01.1E-W0018",
  "manufacturer": "Bosch Rexroth",
  "revision_number": "D",
  "input_voltage": "230V plus or minus 10%",
  "operating_temperature_min": -10,
  "operating_temperature_max": 55,
  "environmental_rating": "IP65",
  "power_consumption": "1.5 kW",
  "specifications": [
    {
      "parameter_name": "Rated output current",
      "value": "9",
      "unit": "A",
      "tolerance": "plus or minus 5%"
    },
    {
      "parameter_name": "Operating frequency",
      "value": "50/60",
      "unit": "Hz"
    }
  ],
  "compliance_certifications": [
    {
      "standard_name": "ISO 9001",
      "certification_body": "TUV",
      "certification_date": "2024-05-11"
    },
    {
      "standard_name": "CE",
      "certification_body": "Self-declared",
      "certification_date": "2025-01-20"
    }
  ]
}

Frequently asked