Read receipts, scans, photos, tables and policy documents with code and AI, and show exactly where every value came from. You build Larkfield's document pipeline, compare the PDF text layer, OCR and a vision model on the same pages, and send only the uncertain values to a person.
Read every lesson as a guest, or sign in to enroll and track your progress.
Enroll to save the lessons that you complete. It is free.
By the end of this course, you will be able to:
Larkfield’s help desk gets documents with its tickets: receipts for warranty and refund claims, some as PDFs from the shop’s till, some as scans and some as phone photos, a few in French. The purchasing team gets supplier price lists and spreadsheets. Today an agent opens every file and types the order number, the date, the total and the items by hand. It is slow, and sometimes the same receipt is typed twice with two different totals.
Grace, the support lead, asks: “Can the system read the receipt, show us exactly where each value came from, and send us only the ones it is not sure about?” In this course, you build that system, doc-extract, one module at a time.
First you look inside the files. A PDF from the till already holds its text; a scan is only a picture of text; a photo is a picture taken at an angle, in poor light. You route each file to the way of reading that fits it, and you compare the PDF text layer, OCR and a vision model on the same pages: what each one gets right, what it gets wrong, and what it costs per page.
Then you keep what makes a value trustworthy: its source. Every value carries its file, page and box on the page, or its sheet and cell, so a person can check it in one click. You validate totals and dates, send uncertain documents to a review queue, record corrections without erasing the original, and measure the pipeline field by field. The documents are made by a program for this course, and every model answer in the lessons comes from a real recorded call.
You completed Building Reliable Applications with LLM APIs and RAG, or you have the same skills. You do not need an account or money. The text layer and OCR run on your computer's CPU, the documents are made by the course's own generator, and every vision-model answer in the lessons was recorded, so the project replays it. A live call with your own key is optional.
Courses to take first: Building Reliable Applications with LLM APIs, RAG: Building AI That Uses Your Data.
People who have built an application on a model API with structured output and a RAG assistant over documents, as in Building Reliable Applications with LLM APIs and RAG, and who now must get reliable values out of PDFs, scans, photos and spreadsheets. Developers, data engineers and AI engineers who automate work with documents.
You can study the ideas of this course without writing code. Read modules 1, 4 and 5 (about 5–7 hours), answer their knowledge checks, and get a free concepts certificate. You can come back for the practice later.
The paths for people who lead or use AI take this course this way. See the paths
Experienced AI practitioners designed this course. AI helped us draft the content; people chose the teaching approach and the examples, and reviewed every lesson. The outputs that you see are the actual results of running the code. How we make our courses
Look inside the files before you read them. Set up the project, tell digital PDFs, scans, photos and spreadsheets apart, see how layout changes meaning, and write the extraction contract that says what Larkfield needs.
Get the text out, the right way for each file. Stop bad files at the door, compare the PDF text layer, OCR and a vision model on the same pages, and prepare images without losing the original.
Keep the structure and the source. Recover sections, a table that spans two pages and a spreadsheet with units, and store the page and box, or the sheet and cell, with every value.
Use a vision model where it helps, and only there. Design its input and schema, combine it with the rules that already work, and keep a missing value missing instead of a plausible guess.
Check before anyone trusts a value. Validate totals, dates and order numbers, decide what goes to review with signals you can measure, and record corrections without erasing the original.
Measure it and hand it over. Score each field by document type, budget pages, tokens and time, and plan reprocessing, deletion and regression sets.
Two end-to-end case studies on real public data. Read a public set of labelled receipts, and recover a multi-page table from a public report and check it against the published data. Complete at least one for the certificate; each one is a portfolio piece.
Complete at least one case study for the certificate. The others are optional. What each one covers
Use every module in one practice project. You collect evidence, and check it against the course outcomes.
Realistic projects from start to finish, on real data, with every output explained. Each one makes a portfolio piece. Complete at least one for the certificate; the others are optional.
Read real receipts2 hr 30 min
Run the course's pipeline on a public set of labelled receipt images, and measure its field accuracy against the published labels.
Extract a table from a public report2 hr
Recover a multi-page table from a public-domain statistics report with its provenance, and check every cell against the agency's published data file.
Loading the reviews…
Took this course? Sign in to review it. Learners can write a review after they complete 25% of the lessons.
