Jev Mini Hackathon
Overview
A taster for our ML engineering programme, run as a mini hackathon. You'll build something on Jev, TypeSafe AI's decision model, in a group of four.
Background
Jev does not generate text. You send it a JSON state and a set of typed questions, and it sends back decisions with probabilities: which of these options, is this true, where on this scale. The questions in one request run in parallel against the same state, so asking 50 questions takes about as long as asking one.
TypeSafe call it a System 1 model, after Kahneman. System 1 is the fast judgement, recognising that a customer message is urgent, and System 2 is working through the recovery plan. A lot of what gets handed to an LLM today is a yes or no, or a choice from a short list, and a model that answers with a calibrated number instead of a paragraph is cheaper, faster and easier to build on.
We'll go through how Jev works and the three question types, and then you'll spend most of the evening building.
Who is this for
Experienced software engineers. Comfortable calling an HTTP API from the language of your choice. No ML experience needed.
What we will cover
Opening talk: what Jev is and what a System 1 model is for
The API: state, questions, and the three answer types (choice, noul, score)
Groups of four: pick a use case, decide why it suits a model that answers with numbers, split the work
Build it, with a TypeSafe API key for each group on the night
What to bring
Your laptop