Jev in Practice: Stock Supply & Demand Analysis in Colab

Course Overview

This course teaches how to implement a narrow, typed judgment step before decision-making using stock research as an example and Jev, TypeSafe’s typed narrow-judgment API, as the interface. It uses fictional, deterministic data representing a synthetic market of four tickers over 300 trading days with invented fundamentals for educational purposes. The course covers building a research pipeline, computing volume proxies, implementing CAN SLIM checks as code, assembling an evidence matrix, defining typed questions, and generating auditable research notes. The second half applies these concepts beyond finance, including PII detection, confidentiality label mismatch detection, browser automation restrictions, and evaluation of API-compatible implementations. An optional chapter demonstrates calling the production Jev API from Colab.

Key Takeaways

• Build a deterministic synthetic four-ticker market in Colab and compute price, relative-volume, and cumulative-volume features without look-ahead bias.
• Implement CAN SLIM checks as code and assemble a seven-letter evidence matrix with pass, fail, and unknown states.
• Define typed Choice, Score, and Noul questions for Jev and answer them through a mock adapter with confidence routing.
• Read reliability diagrams to separate classifier evaluation from investment evaluation and generate auditable research notes.
• Design narrow-judgment flows for PII detection, confidentiality-label mismatch, and permission-limited browser operation.
• Compare local API-compatible implementations with TypeSafe Jev and complete an independent evaluation plan and decision contract.

Prerequisites

• Basic Python knowledge including functions, lists, dictionaries, and some pandas experience.
• A Google account for Google Colab; core notebooks run free and offline.
• General familiarity with stock market terms such as price, volume, and earnings.
• Optional: a TypeSafe Jev API key for running the live chapter, which may incur charges.

Target Learners

• Python developers interested in adding typed, auditable AI judgment steps to data pipelines.
• Individual investors and analysts curious about coding CAN SLIM style checks on synthetic data for study.
• ML and product engineers evaluating narrow-judgment APIs for PII, document labels, or browser automation.
• Learners seeking a rigorous, fiction-based method to practice LLM evaluation without real financial data.