Training Curriculum
LLM-Augmented Text Analysis: A Two-Day Intensive
For researchers, analysts, journalists, educators, and communications staff. No programming background required. Every participant leaves with a finished analysis on their own data.
2
Days, in person — eight 90-minute blocks
0
Programming experience required — spreadsheets are enough
6
Methods, hands-on, on a vetted reference corpus and your own data
1
Finished, written-up analysis of your own corpus by the end of Day 2
The intensive uses Python in Google Colab with modern LLM APIs — but it is built for people who have never written a line of code. Pre-work (about 90 minutes, self-paced) sets up accounts and vets each participant’s own dataset, so the two days are spent analyzing, not troubleshooting.
Day 1 — Foundations and Methods
Taught on a shared, vetted reference corpus so everyone works from the same material and nothing depends on the quirks of anyone’s data — yet.
Workshop Map & Reading Code
See the destination first: one small, live example of every method the workshop covers, run end to end. Then learn to read code rather than write it — open a notebook, load data, change a parameter, re-run. Everyone leaves with the Workshop Map notebook and a one-page handout of the full arc.
Five Paths to Text Data
An honest landscape of how to get text data after the workshop: institutional databases, public and paid APIs (with a live pull into Colab), scraping — including its legal and ethical limits — pre-built datasets, and your own files, PDFs and scanned material included. Everyone leaves with a Data Sourcing Playbook of vetted links and how-tos.
Three Approaches Compared: Lexicon vs. Unsupervised vs. LLM
The same task run three ways on the same corpus — dictionary methods, classic topic modeling, and LLM-based classification — with the strengths, weaknesses, and costs of each laid side by side. This is the block that lets you defend your methodological choices to a reviewer, an editor, or a board. Everyone leaves with the comparison notebook and a “Which Approach When” decision aid.
Structured Extraction & Zero-Shot Classification
The first serious hands-on methods: pull structured information — quotes, entities, dates, claims — out of documents into clean tables, and classify documents into categories you define in plain English. Covers schema design, validation, batching, and what it actually costs. Overnight, participants apply both methods to their own corpus.
Day 2 — Your Own Corpus, Validated
Every method from here on runs on the participant’s own data, with instructor support in the room.
LLM-Labeled Thematic Clustering
Discover what’s actually in your corpus: embed the documents, cluster them, and use an LLM to label each cluster in readable language — ending in a visual theme map of your own data. The modern replacement for classic topic modeling.
LLM-Assisted Qualitative Coding & Claim Extraction
Apply an existing coding scheme at scale or let the data propose codes — with reasoning traces and the researcher’s judgment firmly in the loop. Then extract verifiable claims from a corpus and stage them for follow-up verification, a method built for journalism and policy work.
Validation — Defending Your Analysis
The block that earns the word “rigor”: spot-check sampling protocols, multi-pass agreement, failure-mode reporting, and a reusable methods-section template covering exactly what to disclose — model, prompts, validation, limitations, cost. Without this block, nothing else is publishable; with it, your analysis survives peer review, editors, and skeptical boards.
Next Steps — A Roadmap, Not a Goodbye
Each participant finishes a one-page write-up of what they built, plans their next 90-day project with a structured template, and leaves with a skill-deepening roadmap, a method-extension roadmap (RAG, local models for sensitive data, agentic workflows), a curated reading list, and an alumni community invitation.
Every Participant Leaves With
A finished LLM-augmented analysis of their own corpus
Working notebooks for all six methods, reusable on any dataset
A defensible methods-section template and their own draft
The Data Sourcing Playbook and “Which Approach When” decision aid
A personal 90-day next-project plan
Alumni community access and follow-up office hours
Honest fine print: participants complete about 90 minutes of pre-work, bring their own dataset (we vet it in advance and provide a substitute if needed), and spend roughly $5–15 of their own API credit during the two days. We tell you this up front because that’s how the whole workshop runs.
Bring this to your team
The intensive is delivered for organizations and cohorts — newsroom teams, research groups, agencies, and association memberships. Shorter single-day and half-day formats are available, built from the same modules.
Ask about training