Data Science & AI Tech Talks - in-person meetup at Ripjar's Eagle Tower offices

This is an in-person meetup hosted by Ripjar in their Eagle Tower offices. The outline schedule for the evening is:

6pm - arrival, drinks & pizza

630pm-730pm - Tech Talks

730pm-8pm - drinks, pizza & networking

Optional - retire to a local pub to continue the conversation

We have two Tech Talks:

*** "Using AI to Build AI"

Dan Saunders & team (Ripjar)

AI is no longer just the subject of data science—it’s becoming a collaborator in the practice itself. From generating gold-standard datasets at scale, to zero-shot classifiers that bypass months of manual labeling, to automated QA “judges” that enforce guardrails and validate outputs, AI is reshaping the data science workflow. It even writes code, builds models, and (ironically) drafts talk abstracts. These tools bring dramatic speed-ups, but also raise new challenges: hidden inefficiencies, higher computational costs, and the temptation to cut corners in validation. In this talk, Dan and his team will showcase recent initiatives at Ripjar that have put “AI to build AI” into practice—highlighting both the wins and the pitfalls when machines start shaping how we design, test, and deliver data science solutions.

*** "Test Driven Data Analysis for Books and Other Documents"

Nick Radcliffe (Stochastic Solutions & University of Edinburgh)

Test-Driven Data Analysis is a methodology and supporting software for improving the quality of data and of data analysis processes. There is software support for automatic constraint generation and the use of those constraints (usually, as adapted) for validating data. There is also software support for writing reference tests for analytical processes and pipelines. This includes not only custom analytically-appropriate assertions for Python’s unittest and pytest, but also a pre-GenAI form of AI-driven test generation for command-line programs and scripts in any language (Gentest).

The speaker has a forthcoming book (Test-Driven Data Analysis, Nicholas J. Radcliffe, CRC Press, in press) covering these topics and other forms of errors in analysis, including errors of interpretation, errors of process, errors of applicability and errors of judgement. He is in the process of treating the entire book as an analytical process to which TDDA can be applied, giving rise to the notion of Test-Driven Document Development (TDDD). This is can be seen as an extension of computational documents (RMarkdown, Quarto, Jupyter etc.) that uses tests to reduce or eliminate the danger of co-rusting, whereby the code and results in a computational document drift away from correctness together without actually breaking.

The talk will begin with an overview of TDDA, and then show how this can be applied to the specific content of document creation, including book-length documents.