# Glimps technology > More than simple OCR. We use hybrid AI models to understand financial context, structure, and semantics. Source: https://useglimps.com/en/technology | Language: en | Updated: 2026-07-06 This Markdown file is the plain-text twin of the page above, generated at build time from the same render. --- More than simple OCR. We use hybrid AI models to understand financial context, structure, and semantics. Read: text and tables, scans included ### Multimodal OCR Extracts text, layout, and handwriting simultaneously. ### Semantic Mapping Maps descriptions to GL codes via vector embeddings. ### Continuous Learning Every correction is carried to that supplier's next invoice. ## Semantic Understanding Traditional OCR relies on keywords and templates (e.g., looking for "Total:" at bottom right). This breaks easily with varying layouts. Glimps understands the meaning of words, not just their position. It knows that "MacBook Pro" is conceptually similar to "Dell XPS" and should be booked as Hardware, even if it has never seen the specific invoice layout before. A new line: "MacBook Pro", from an invoice Glimps has never seen A bold number at the top right: most likely the date ## Visual Layout Analysis Our models see document structure. They understand that a bold number at the top right is likely the date, and that grids imply tables. We analyze the entire document layout visually. This allows us to successfully extract data from complex tables, multi-column layouts, and even handwritten notes that confuse standard OCR. ## You decide what Glimps posts by itself By default every invoice is read, coded and ready, and you press post. Per supplier you can let Glimps post by itself. Eight checks always stop a posting, such as a possible duplicate or a different bank account. - Post automaticallyOnly for suppliers you release - Prepared for youRead and coded, you press post Glimps gives every field a confidence score ## Frictionless Review Interface For every suggested GL account you see why Glimps picked it: an earlier posting, a rule or a suggestion. If it is wrong, you change it in one click. The ledger account is sure, the amount is not: only that field lights up Glimps suggests a ledger account: 4600 General expenses ## Human Feedback Loop When you correct a GL account, Glimps takes it straight to the next invoice of that supplier. That learning stays inside your own administration. ## PDF, scan, photo and UBL From crumpled receipts to high-quality PDFs. Glimps unifies your input streams. ### Every common file PDF (digital and scanned), JPG and PNG, and a UBL file together with its PDF ### Several invoices in one PDF Glimps notices and asks whether to split the PDF. ### A second read when in doubt If the sum does not add up, Glimps reads the document again more closely. ## Technology FAQs Questions about our AI engine and capabilities. Want to try Glimps yourself?Try interactive demo Does Glimps work with handwritten invoices? Often, yes. Glimps also reads scans, photos and handwritten receipts. The worse the image, the more often Glimps asks you to take a look, and the math check catches reading errors before anything is posted. How long does it take to learn a new vendor layout? Zero time. Glimps uses semantic understanding (like a human), not templates. It works instantly on layouts it has never seen before. Can I process receipts as well as invoices? Absolutely. The system automatically detects document type and extracts relevant fields for receipts, credit notes, and invoices. What happens if the AI is unsure about a field? Then Glimps does not post by itself. The invoice is ready with what Glimps read, and where the sum does not add up or something is missing, Glimps says exactly what. Your correction is carried to that supplier's next invoice. Yes, Glimps extracts full line item details including descriptions, quantities, unit prices, and SKU codes, even from complex tables spanning multiple pages. How does semantic mapping work for GL accounts? Glimps converts text descriptions into vector embeddings to understand context. For example, it knows that "Uber" and "Taxi" are semantically similar and maps both to your "Travel Expenses" ledger automatically.