Module 04
Real-World Text Tasks
Put prompting to work on the text jobs real businesses pay for: summaries, translations, tone fixes, ticket routing, code help, training bots and support assistants. Each lesson follows one Crumb problem from a naive demo to something dependable, with code checks and measurements. Two projects tie it together: a tested utilities package on the OpenAI API and a customer support agent that knows when to hand over to a human.
By the end of this module you can
- write summarisation, translation and tone briefs that name the reader, the length and what must not change, and check faithfulness in code
- build a ticket router with a structured triage form and code-level escalation rules, and measure it with precision and recall
- use LLMs to generate, modify and explain code safely, with real context, diffs, read-only access and tests
- design role-play bots, learning advisors and grounded support assistants with character sheets, hidden state and separate coaching calls
- ship a production-style Python package and CLI with config, retries with backoff, validated outputs and offline tests
- build a support agent that classifies, retrieves help articles, drafts cited replies and escalates to a human when out of its depth
Lessons
6 in this module- 01Summarisation, translation and tone controlTurn summaries, translations and tone rewrites from demos into dependable tools by naming the reader, fixing what must not change, and checking numbers and protected terms in code.
- 02Support ticket routing and analysisRoute support messages with a multi-label triage form, enforce escalation rules in code, measure precision and recall against a labelled set, and turn the results into weekly analysis.
- 03Generating, modifying and explaining codeGet useful SQL and code from an LLM by supplying the relevant schema and house rules, asking for diffs and assumptions, running results read-only, and checking them against numbers you trust.
- 04Role-play bots, learning advisors and assistantsBuild persona-driven bots that work: a training simulator with a character sheet and hidden state, a Socratic tutor with a hint ladder, and a support assistant grounded in your own handbook.
- 05Project: Working with the OpenAI APIShip crumbkit, a tested Python package and CLI on the OpenAI API for summarisation, translation, tone control, ticket routing and code help, with shared config, retries with backoff and offline tests.
- 06Project: Customer support agentBuild a Crumb support agent that classifies messages, looks up orders in code, retrieves help articles with BM25, drafts cited replies, and escalates to a human whenever it is out of its depth.