Subscribe
Sign in
Koncile
p/koncileCustomisable OCR for all your data extraction needs
Start new thread
Jules Ratier

Koncile - Customisable OCR for all your data extraction needs

by•

📄 Effortless data extraction from your PDF invoices, quotes, and more 💬 Just type the fields you need 👌🏼 Get all your line items and tables in a neatly structured format 👀 Powered by computer vision + LLM, outperforming standard OCR systems

Add a comment

Replies

Best
Jules Ratier
Hunter
📌
Hi Hunters! I’m Jules, cofounder at Koncile.ai 🚀 Koncile is a fully customizable OCR that combines computer vision with LLMs to extract custom data from your PDFs. The tool is specifically designed for invoices, quotes, purchase orders, or any document containing recurring line items or tables. When we started this project, I was searching for an OCR that could accurately extract line items. I tested over 40 off-the-shelf solutions on the market, but I was consistently disappointed by the results: ❎ Inaccurate data due to multiple errors ❎ Poor table extraction ❎ No way to create custom fields (I wanted to consistently capture those EANs) So, we built Koncile Extract to deliver stellar extraction results ✨ and complete customization 🖊️. Our secret sauce? We leverage the power of open-source computer vision models—great for recognizing characters—paired with LLMs’ vision input, which excels at understanding layouts and classifying data. The process is ultra-simple: 1️⃣ Create your extraction schema or start with one of our 100+ templates 2️⃣ Test it on a sample file 3️⃣ Download the data in Excel format or integrate it via our API We’re offering a free 30-day trial account, so you can give it a try: https://app.koncile.ai/auth/regi...
Raju Singh
@jules_ratier Hey Jules, Congrats on the launch. DO you have specific false positives / negatives data for how accurate your model is wrt OCR. Any specific language being benchmarked here?
Jules Ratier
Hunter
@imraju We use a combination of traditional optical recognition technology w/ LLM. We've got really high success rate (+99%) for standard data that companies are willing to extract in invoices, quotes or tables, including line items
Steven Renwick
what scale can this run at? could you process hundreds of pdfs in parallel?
Jules Ratier
Hunter
@major_grooves Exactly, you define the field to extract on a document type (invoice, quote, etc.), and after run it on multiple files at the same time
Henri de Bouteiller
Wonderful software that helped our company saved 100ks$ !!! Great work team
Bon
Congrats on the launch! I find that the accuracy of OCR in my daily use is quite good, and I haven't encountered many significant issues. What are the advantages of LLM-driven OCR? Also, does it slow down compared to traditional solutions, and will ti be more expensive?
Huzaifa Shoukat
Congrats on the launch! 🎉 This OCR tool looks amazing. How do you handle multi-page docs with varying layouts?
Jules Ratier
Hunter
@ihuzaifashoukat Yes, the tool handles long documents! By definition, the goal is to capture information with any type of layout. With @tristan_thommen, we've been dealing for a year with the ugliest invoices ever to refine the extraction quality ;) If you have several documents in a single file, you can put specific instruction to capture only the info you need.
Max Comperatore
Launching soon!
Hell yes, data extraction with natural language? Im all in. Upvoted.
Germán Merlo
Wow Jules! I'm obsessed on data extraction tools and hacks, and Koncile looks super interesting on that. I'm sure many founders will take the most of it and wish you all the best here!
Munna Aziz
Congrats, Jules and the Koncile team! 🎉 Koncile sounds like a fantastic solution for anyone dealing with detailed document data extraction, especially those tricky line items in invoices and purchase orders. Love how it combines OCR and LLMs for more accurate, customizable results—especially for fields like EANs!