OCR API Scans and Photos to Markdown

• 

  # INVOICE No. 2026/11/0457
  
  ## Seller:
  Nordwind Tools GmbH
  Hafenstraße 12, 20457 Hamburg
  VAT ID: DE 298 114 551
  
  Issue date: 2026-11-02
  
  ## Buyer:
  Kowalski & Syn sp. z o.o.
  ul. Żółkiewskiego 7/3, 31-539 Kraków
  NIP: 676-25-81-904
  
  Due date: 2026-11-30
  
  <table>
    <thead>
      <tr>
        <th>#</th>
        <th>Item</th>
        <th>Qty</th>
        <th>Unit price (EUR)</th>
        <th>VAT</th>
        <th>Net (EUR)</th>
      </tr>
    </thead>
    <tbody>
      <tr>
        <td>1</td>
        <td>Torque wrench TW-40, 1/2"</td>
        <td>12</td>
        <td>84.50</td>
        <td>0%</td>
        <td>1,014.00</td>
      </tr>
      <tr>
        <td>2</td>
        <td>Socket set S-94 (94 pcs)</td>
        <td>5</td>
        <td>129.00</td>
        <td>0%</td>
        <td>645.00</td>
      </tr>
      <tr>
        <td>3</td>
        <td>Shipping DHL Freight</td>
        <td>1</td>
        <td>48.90</td>
        <td>0%</td>
        <td>48.90</td>
      </tr>
    </tbody>
  </table>
  
  Total net: **1,707.90 EUR**
  Reverse charge, Art. 196 Directive 2006/112/EC

• 

  # MERCADO LUMEN
  
  Calle Mayor 14, 28013 Madrid  
  NIF B-86512034  
  Tel. 915 22 41 70
  
  ---
  
  Factura simplificada T-0415/2611  
  Fecha 02/11/2026 Hora 18:42  
  Caja 3 Atendido por: Lucía
  
  ---
  
  <table>
    <thead>
      <tr>
        <th>Item</th>
        <th>Cantidad</th>
        <th>Precio Unitario</th>
        <th>Total</th>
      </tr>
    </thead>
    <tbody>
      <tr>
        <td>Pan de masa madre</td>
        <td>1</td>
        <td></td>
        <td>3,20</td>
      </tr>
      <tr>
        <td>Aceite oliva v.e. 1 l</td>
        <td>1</td>
        <td></td>
        <td>9,85</td>
      </tr>
      <tr>
        <td>Tomate rama kg</td>
        <td>1,250</td>
        <td></td>
        <td>3,11</td>
      </tr>
      <tr>
        <td>Queso manchego cur.</td>
        <td>1</td>
        <td></td>
        <td>7,40</td>
      </tr>
      <tr>
        <td>Agua mineral 1,5 l</td>
        <td>6</td>
        <td></td>
        <td>3,30</td>
      </tr>
      <tr>
        <td>Café molido 250 g</td>
        <td>2</td>
        <td></td>
        <td>6,58</td>
      </tr>
    </tbody>
  </table>
  
  
  ---
  
  TOTAL EUR 33,44  
  Tarjeta **** 4417 33,44
  
  ---
  
  <table>
    <thead>
      <tr>
        <th>IVA</th>
        <th>Base</th>
        <th>Cuota</th>
      </tr>
    </thead>
    <tbody>
      <tr>
        <td>4%</td>
        <td>13,18</td>
        <td>0,53</td>
      </tr>
      <tr>
        <td>10%</td>
        <td>17,94</td>
        <td>1,79</td>
      </tr>
    </tbody>
  </table>
  
  
  ---
  
  ¡Gracias por su visita!

• 

  # Cold-chain losses in regional food distribution
  
  Working paper 14/2026 · Logistics Research Unit
  
  ## 1. Introduction
  
  Perishable goods lose value at every hand-off between warehouse, truck and shop. In the 2025 audit of 212 deliveries, temperature excursions above 8 °C lasted on average 37 minutes per route, most of them during unloading. This paper measures how much product those excursions cost and which routes account for most of the loss.
  
  We combine data loggers on 64 trucks with invoices from 18 retailers. Each delivery is matched to its loss report within 48 hours, which lets us separate spoilage in transit from spoilage on the shop floor.
  
  ## 2. Data and method
  
  Loggers recorded the air temperature every 60 seconds. A delivery counts as an excursion when the reading stays above 8 °C for more than five consecutive minutes. Losses are valued at the retailer’s purchase price, net of VAT.¹
  
  Table 1 summarises the sample by product group. Dairy and fresh meat make up 61% of the value shipped but 78% of the value lost.
  
  <table>
    <thead>
      <tr>
        <th>Product group</th>
        <th>Deliveries</th>
        <th>Excursions</th>
        <th>Loss (EUR)</th>
      </tr>
    </thead>
    <tbody>
      <tr>
        <td>Dairy</td>
        <td>74</td>
        <td>19</td>
        <td>4,210</td>
      </tr>
      <tr>
        <td>Fresh meat</td>
        <td>51</td>
        <td>16</td>
        <td>6,980</td>
      </tr>
      <tr>
        <td>Fruit and vegetables</td>
        <td>63</td>
        <td>9</td>
        <td>1,345</td>
      </tr>
      <tr>
        <td>Frozen</td>
        <td>24</td>
        <td>3</td>
        <td>610</td>
      </tr>
      <tr>
        <td><strong>Total</strong></td>
        <td><strong>212</strong></td>
        <td><strong>47</strong></td>
        <td><strong>13,145</strong></td>
      </tr>
    </tbody>
  </table>
  
  *Table 1. Deliveries, excursions and losses by product group, March–October 2025.*
  
  ## 3. Results
  
  Routes with more than six stops had twice the excursion rate of shorter routes (31% against 15%). The effect holds after controlling for outside temperature and truck age. Unloading with the doors open for longer than ten minutes explains about half of the difference.
  
  Moving the two longest routes to an early-morning slot would have cut excursions by an estimated 40% at no extra fuel cost. We discuss the limits of this estimate in Section 4.
  
  ## 4. Limitations
  
  Retailers report losses by hand and may round them. Loggers sit near the doors, so they overstate the temperature in the middle of the load.
  
  ---
  
  ¹ Purchase prices come from the invoices; where a line item was missing, we used the median price for that product in the same month.
  
  7

• Finance

  Invoice and receipt capture

  Line items arrive as an HTML table, so your code reads rows and columns straight from the markup. Check totals and VAT IDs in your own code before you book anything.

• RAG

  Scanned PDFs into a search index

  Headings stay Markdown headings and tables stay tables, which gives you clean chunk boundaries. Render each PDF page to an image first; the API takes images only.

• Reports

  Numbers out of charts

  Feed slides and PDF reports through it and each bar chart arrives as a small table of figures, so the numbers reach your spreadsheet without retyping. The chart test is on the LightOnOCR-3 page.

• Research

  Papers with math

  Formulas come back as LaTeX, inline ones as $…$ and display equations as $$ blocks that KaTeX or MathJax render.

Prices from the deAPI /price endpoint, 3 November 2026. Other sizes and the per-megapixel rate are in the full price table. Before a batch, POST its width and height to /api/v2/images/ocr/price for an exact quote.

1. 1

   Render the pages

   Rasterise every page to PNG; pypdfium2 and pdftoppm both work. LightOn recommends 2048 px for the longer side, and at half that size one figure on our test page came back wrong.

2. 2

   Send one request per page

   POST each image to /api/v2/images/ocr with model LightOnOcr_3_4B_Q8. Quote the batch first with the /price endpoint.

3. 3

   Join the Markdown

   Read each result from the job or your webhook and join the pages in order.