Course 3.13 · Curriculum 2026.1
Vision, documents, and the confidently wrong answer
Read an image or a document without inventing what is not on it. A blank business card once produced a confident name and a confident company, and that single incident is the reason every rule in this course exists.
- Estimated effort
- 3 hr
- Lessons
- 26
- Modules
- 5
- Material
- 3 hr
Estimated effort is the sum of the lesson and lab times in the course outline, for this version of the course.
This course is planned and not on sale. The curriculum below is the plan, and it may change before the course opens.
See the courses on sale01What you will learn
6 capabilities, stated as things you can do
Each one is what you should be able to do at the end, not what the course covers. If an outcome is not testable, it is not an outcome.
- Force image reading through a real vision call, rather than a chat window that may never receive the pixels.
- Set temperature to zero against a strict schema, and say what each of those two buys you.
- Design a schema whose fields can come back null, so absence is representable.
- Catch a confident fabrication by requiring the evidence span beside the value.
- Route a low-confidence extraction to a human queue instead of to a record.
- State the rule out loud: a confident wrong answer is far worse than "I could not read that".
02Curriculum
5 modules, 26 lessons
Every lesson shows its length and its type, from the course outline.
5 modules · 26 lessons · 3 hr of material
A blank business card once produced a fully fabricated name and company. The incident that every rule in this course traces back to.
- VideoA blank business card, and the name it confidently inventedPreview6 min
- VideoWhy the model answered at all, when there was nothing to read6 min
- VideoThe agent does not reliably receive full-resolution pixels6 min
- ReadingWhat “confidently wrong” costs against “I could not read that”5 min
- VideoThe standing rule this incident produced5 min
Never transcribe a pasted image by looking at it. Card and document reading is forced through a real vision call at temperature zero.
- VideoNever transcribe a pasted image by looking at it6 min
- VideoTemperature zero, and what it buys you6 min
- VideoA strict JSON schema, and what it buys you separately6 min
- ReadingTwo guardrails, two different jobs: determinism and shape5 min
- LabLab: force a card read through a real vision call17 min
A schema whose fields can come back null, so absence is representable, and an evidence span beside every value so a fabrication has nowhere to hide.
- VideoDesigning a schema whose fields can come back null6 min
- VideoWhy “unknown” has to be a valid answer, not a missing one6 min
- VideoRequiring an evidence span beside each value6 min
- VideoCatching a fabrication because the evidence span does not exist6 min
- ReadingA schema review checklist for vision output5 min
- LabLab: build a nullable schema and reject an answer with no evidence15 min
A low-confidence extraction goes to a human queue, never to a record. And one document type is blocked from upload entirely.
- VideoRouting a low-confidence extraction to a human queue instead of a record6 min
- VideoBlocking upload entirely for one document type6 min
- VideoRouting a blocked document to a single address instead6 min
- ReadingDeciding which document types get blocked before the first upload5 min
- LabLab: build the confidentiality guard and prove it blocks15 min
Comparing a declared address and time against a GPS trail, surfaced as a correction banner rather than a silent overwrite.
- VideoComparing a declared address and time against a GPS trail6 min
- VideoHaversine distance, and the plus-or-minus 60 minute window6 min
- VideoA correction banner with an apply-GPS-time action6 min
- ReadingFive cases verified end to end, and what each one caught5 min
- QuizCheck: vision and the confidently wrong answer7 min
04Before you start
What you need first
Assumed knowledge and setup
- Course 3.6. Output validation is the backstop for everything here.
- A model endpoint with a vision path. The stub from course 3.4 carries one.
- Sample documents are supplied, including the blank card.
Before this course
05Who teaches it
Elliot Saha
Co-founder, Chief Technology Officer
Platform architecture, custom Apex and Lightning Web Components, AI systems, and the Academy platform itself.
06Reviews
No learner reviews yet
This course has not been taken by enough people to publish an honest rating, and we will not print invented quotes on a page that sells verification discipline. Here is what we can evidence instead.
- Assessment
- Lesson checks onlyThis course carries lesson checks and issues no certificate.
- Curriculum version
- 2026.1You enrol into a version. Content changes do not move you mid-course, and your required-lesson count is snapshotted at enrollment.
- Refunds
- 14 daysAn individual purchase can be refunded within 14 days if you have completed less than 20% of the lessons. Team seats are refundable before they are assigned. From the Academy terms.
Start 3.13
This course is in development and not yet open for enrolment. The courses open now are listed in the catalogue.
- 3 hr estimated effort, labs included
- A ShareCo certificate is not a Salesforce certification