SoceTonAI DoxTract
Draw Once. Extract Alltime.
Businesses spend countless hours manually entering data from invoices and other business documents into their systems. Existing OCR solutions can be expensive, require technical setup, or struggle with consistent extraction across recurring document layouts. We built SoceTonAI DoxTract to make document data extraction simple, affordable, and reusable through templates. Create a template once, then extract structured data from similar documents in seconds—without training custom AI models.
⚙️ Benchmark results
𝟭𝟬𝟬% 𝗔𝗰𝗰𝘂𝗿𝗮𝘁𝗲 𝗜𝗻𝘃𝗼𝗶𝗰𝗲 𝗧𝗲𝗺𝗽𝗹𝗮𝘁𝗲 𝗗𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻 — 𝗮𝘁 𝗡𝗲𝗮𝗿-𝗭𝗲𝗿𝗼 𝗖𝗼𝘀𝘁
Benchmarked SoceTonAI DoxTract — a template-driven document extraction system — on a real-world dataset of 7,000 invoice images (Kaggle Dataset), using 4,000 invoices across 4 distinct templates.
𝐖𝐞 𝐀𝐜𝐡𝐢𝐞𝐯𝐞𝐝 𝟔𝟑% 𝐭𝐨 𝟏𝟎𝟎% 𝐈𝐧𝐯𝐨𝐢𝐜𝐞 𝐃𝐚𝐭𝐚 𝐄𝐱𝐭𝐫𝐚𝐜𝐭𝐢𝐨𝐧 𝐀𝐜𝐜𝐮𝐫𝐚𝐜𝐲 (𝐓𝐞𝐦𝐩𝐥𝐚𝐭𝐞-𝐃𝐞𝐩𝐞𝐧𝐝𝐞𝐧𝐭) 𝐚𝐭 $𝟎.𝟔𝟎 𝐩𝐞𝐫 𝟏,𝟎𝟎𝟎 𝐏𝐚𝐠𝐞𝐬
Each template was tested independently using the 𝗳𝗿𝗲𝗲𝗺𝗶𝘂𝗺 𝗺𝗼𝗱𝗲𝗹 (𝘀𝗺𝗮𝗹𝗹𝗲𝘀𝘁 𝗺𝗼𝗱𝗲𝗹).
📊 Full results + analysis:
https://sadidul.medium.com/100-accurate-invoice-template-detection-at-near-zero-cost-3138c923ce57
⚙️ How it works
Draw template (map fields visually)
Save template to cloud
Upload invoices → get structured JSON instantly
📚 Docs
Data extraction: https://soceton.com/docs/doxtract-data-extraction
Template basics: https://soceton.com/docs/doxtract-template-basics
Create template: https://soceton.com/docs/doxtract-create-template
🆓 Free tier
5 templates
20 pages/month
Cloud template storage
Feedback welcome — especially on accuracy edge cases + pricing model.
Product: http://soceton.com/doxtract
1 Comment
The interesting part here isn’t the extraction accuracy itself.
It’s whether templates become the long-term product abstraction, or just an onboarding mechanism before a different model takes over.
That decision tends to shape both pricing and who actually ends up using the product long-term.