Browse categories
English
$
USD
Connect to your AI
I will build a tested ai tool to extract data from emails, pdfs and messy text
Bartosz S.
About this gig
Emails, PDFs, web pages, notes, reviews: the information you need is in there, but as free text. Copying it into a spreadsheet by hand takes hours, and mistakes still slip through. I build you a tool that does this work.
The tool reads your texts and fills the fields you choose, sorts each one into your categories, or checks a document against your list of required items as found, missing or unclear. You get the results in Excel, CSV or JSON.
Why you can trust the result:
- each value shows the passage it came from
- your rules are applied, such as totals that must add up or allowed values
- values that break a rule or have no source go to a review list
- a final test on files kept aside measures accuracy per field, category or item
Basic is a test on 20 of your files. Standard and Premium include the tool, run notes and a test report, on cloud AI or a local model, as your data allows.
Past work: for a bookkeeping startup, a desktop app that pulls IDs, dates and VAT amounts from the OCR text of scanned Hebrew invoices into Excel, checked by a check digit and VAT arithmetic, extended over 1.5 years.
Get to know Bartosz S.
Bartosz S.
Web Scraping, Data Pipelines, ML and LLM Tools
- FromPoland
- Member sinceAug 2019
- Avg. response time1 hour
Languages
English, Polish
I start by asking why: what the goal is, how success will be measured and whether the data can support it. Then I design the process and build the tools to run it, from collecting data on the web and from internal systems to the reports people decide on, and I keep them running.
Web scraping since 2016, including sites behind anti-bot protection, recurring datasets delivered monthly, data pipelines in Python and SQL, dashboards, and machine learning and LLM tools tested on real cases. Working with data since 2011.
My Portfolio
FAQ
Which documents can it read?
Files you supply: text PDFs, Word files, saved e-mails, saved web pages and plain text, up to 5 pages each and in one language. Scans need their text recovered first (OCR); I quote that separately after seeing your scans.
How accurate is it?
Basic shows feasibility, not a guaranteed accuracy. For Standard and Premium we agree answers you approve, an acceptance threshold and what happens if it is missed, before ordering. The report shows accuracy per field, category or item; spot checks remain useful.
Do my documents leave my computer?
Only if you allow it. If you allow cloud processing, the tool uses a cloud model chosen for your case; if not, it runs a local model on your computer. Samples you send me do leave your computer; without your explicit permission I do not upload them to a cloud model.
What do I need to run it?
It depends on the model and the length of your files. Send me your system, memory and graphics card before ordering; I confirm it fits and measure the processing time on your files. Hardware and hosting are not included.
Why not a ready tool like Dext or Hubdoc?
Try them first if they meet your needs. A custom process makes sense when your files fail their trial, you need your own checklist, or processing must stay local. Basic tests whether it is worth it.
Can it tell me whether a document is compliant?
It shows which required items are present, with quotes, and which need a person to look. The decision stays with you.
What if my documents change later?
A new layout, new fields or a new model version means re-running the test set. Updates are a paid order or a monthly subscription.
What do revisions cover?
Fixes against the agreed fields, items and rules. New document types or fields are an upgrade or a new order.
