Field guide

Eleven tasks written up in depth. Each one covers the mechanism, the sub-families of model and when each is the right pick, use cases in a large bank with the part that makes them hard, working code, and the things that catch people out.

Text

Text classification

One piece of text in, one label out.

6 model families 5 bank use cases 4 snippets

Text

Named entity recognition

A label for every word, not for the document.

4 model families 4 bank use cases 3 snippets

Text

Zero-shot classification

You supply the categories at query time. No training.

3 model families 3 bank use cases 2 snippets

Text

Question answering

Point at the span of text that answers the question.

3 model families 3 bank use cases 2 snippets

Text

Summarization

Long document in, short document out — and possibly invented facts.

4 model families 4 bank use cases 3 snippets

Text

Text generation and LLMs

Predicts the next token, repeatedly. Everything else is emergent.

4 model families 4 bank use cases 3 snippets

Text

Embeddings and retrieval

Text becomes vectors, and meaning becomes distance.

4 model families 4 bank use cases 3 snippets

Documents and vision

Document QA

Read a scanned page, where position on the page carries meaning.

4 model families 4 bank use cases 2 snippets

Audio

Speech to text

Audio in, transcript out — and the hard part comes next.

4 model families 4 bank use cases 2 snippets

Tabular and time series

Tabular classification

Rows and columns in, a category out. Where most bank ML actually lives.

4 model families 4 bank use cases 3 snippets

Tabular and time series

Time series forecasting

History in, future out — and the evaluation design is the hard part.

4 model families 4 bank use cases 3 snippets

Tasks without a written guide are omitted rather than stubbed. They are still browsable under Browse models.