Hassan K.

OCR, document data extraction, and document fraud detection. One client hired me five times to build their fraud prevention product. Five contracts, 5.0 stars every time. Most document projects fail in the same place. The happy path works fine, then a real customer uploads a blurry phone photo of a receipt, or a scan at an angle, or a PDF that is really just an image, and accuracy falls apart. I build for that part first. WHAT I BUILD Invoice OCR, receipt OCR, and document data extraction into clean structured JSON Document AI pipelines for classification, parsing, field validation and export Intelligent Document Processing (IDP) workflows for invoices, receipts, IDs, passports, forms and contracts Identity document verification and identity document fraud detection Document tamper detection using pixel level image forensics Image quality scoring that rejects unusable uploads before they reach verification Computer vision models for detection, recognition and quality inspection REST APIs so the output drops straight into the stack you already have WHERE I AM DIFFERENT Anyone can point a cloud OCR service at a clean PDF. The hard part is knowing when the output is wrong, and knowing when a document has been edited before it was ever uploaded. I built an identity document fraud detection engine using Error Level Analysis, OpenCV and C# that flags manipulated IDs and returns per region confidence scores. I built a document tamper detection API that scores files across multiple independent forensic signals and returns a JSON verdict with the evidence attached. I built an image quality assessment module that grades every upload on blur, glare, exposure and framing before it enters a verification pipeline, so bad inputs get caught at the door instead of quietly producing confident wrong answers. Same client for all of it, four separate contracts, because the accuracy held up. If you work in fintech, lending, insurance, remittance, background checks, KYC onboarding, or any marketplace that has to verify who its users are and whether their documents are real, this is the exact problem I spend my time on. BEYOND EXTRACTION Reading the document is only half the job. Something has to happen next. I build the document automation around it, connecting extraction pipelines to CRMs, databases, accounting systems and internal tools with Python, so the output moves through your business instead of sitting in a folder waiting for someone to look at it. That includes LLM work over extracted text where it earns its place, and document gathering when the files have to be collected before they can be read. CORE STACK Python, OpenCV, Tesseract OCR, PaddleOCR, C#, .NET, AWS Textract, Azure Document Intelligence, Google Document AI, OpenAI and other LLM APIs, REST API development, AWS and GCP deployment HOW I WORK I tell you what will not work before you pay for it. If a cheap third party API solves your problem instead of a custom model, I will say so and you can decide. Clients come back because the accuracy is real, the timelines are real, and I raise problems early instead of at delivery. Send me three or four sample documents, including your worst ones, and tell me what you need out of them. I will tell you straight whether it is a two week build or a two month one, what accuracy you can realistically expect, and what it will take to get there.