FinishedUniversityCase study

AI Workflow DSL

A Python implementation of a DSL for multi-agent workflows: a lexer, grammar with generated parse tables, parser, CLI, and documentation generator, exercised against suites of valid and invalid sample programs.

Problem

Describing how multiple AI agents hand work to each other in plain code buries the workflow structure in boilerplate — the course challenge was to design a small language where the workflow itself is the program, and to build the tooling to parse it.

Solution

The project defines the grammar and generates parse tables from it, then implements a lexer and parser that accept a corpus of valid DSL programs and reject invalid ones with useful errors. A CLI runs programs end-to-end and a doc generator keeps the written deliverables in sync with the grammar, verified by a unittest suite.

What I learned

Generating the parse tables from the grammar instead of hand-writing them meant every grammar change was cheap — the hard design work was in the language, and the tooling followed mechanically.