8Course 8Machine learning & AIAdvanced
Applied AI for Analysts
Use large language models in real analysis work: extraction, classification, drafting, and code, with controls that hold up.
- Chapters
- 9
- Time
- 8 hours
- Format
- Self-paced
- Tools
- Any major LLM API or chat product; Python examples provided
What you will be able to do
- Decide which analysis tasks suit a language model and which stay deterministic.
- Extract structured fields from free text, documents, and web pages with validation.
- Classify and categorise at scale, with a labelled sample to measure accuracy.
- Draft investigation notes, summaries, and stakeholder updates from structured case data.
- Handle personal data under India's DPDP Act, log what was sent, and review before acting.
Chapters
9 chapters. Chapter 1 is free below.
- 1
Where a model fitsFree to read
A task taxonomy: extract, classify, summarise, draft, code. Failure modes of each.
- 2
Prompts as specifications
Writing the instruction, the format, and the examples. Versioning prompts like code.
- 3
Extraction
Free-text fields, PDFs, and web pages to structured rows. Validation and confidence.
- 4
Classification at scale
Categorising tickets, merchant descriptions, or dispute reasons; measuring against a labelled sample.
- 5
Drafting
From case record to note, summary, or email. The review checklist before sending.
- 6
Models as a coding assistant
SQL and pandas with a model beside you, and how to check what it wrote.
- 7
Controls
Data minimisation, PII handling under the DPDP Act, logging, and human review.
- 8
Evaluation
A small labelled set, a scoring rubric, and a decision to ship or stop.
- 9
Capstone: an extraction and triage pipeline
Free-text merchant applications to structured risk fields, with accuracy measured.
Free chapter 1 of 9
Where a model fits
Language models are good at tasks where the input is messy language and the output is language or a simple structure: pull the company name and address out of this paragraph, decide which of these eight categories this complaint belongs to, summarise these twelve notes into one. They are bad at tasks that need exact arithmetic, exact recall of your data, or an answer that must be right every time. Summing a column is a SQL job. Checking whether a GSTIN exists is an API job. Neither should go to a model.
The test for a good fit has three parts. The task is done today by a person reading and typing. There is a way to check the output, either automatically or by sampling. And a wrong answer is recoverable: it goes to review, not straight to a customer or a ledger.
Most analysis teams have a pile of exactly these tasks: free-text fields nobody has categorised, PDFs nobody has keyed in, weekly summaries somebody writes by hand. Clearing that pile is worth more than any ambitious project, and it teaches the team what the models get wrong on their own data.
The chapter ends with an exercise: list ten repetitive tasks in your team, score each against the three-part test, and pick two. Those two become your projects for the rest of the course.
Who it is for
Analysts and team leads who want to use language models on real work, from cleaning messy text fields to drafting investigation notes, without leaking data or trusting output blindly.
Before you start
Comfortable with spreadsheets and basic SQL. Python helps for the pipeline chapters but the concepts do not need it.
Taught by
Paul Montero
Founder and instructor
Questions about this course
How long do I have access?
Is the dataset included?
Can I get a refund?
incl. GST, 7-day refund