Перейти к содержанию

ADR 0001: AI-assisted engineering workflow

Status

Accepted

Context

The course is developed as a reproducible engineering route for SDR education:

theory -> modeling -> fixed-point -> HDL/FPGA -> RF frontend -> TX/RX -> synchronization -> IQ recording -> electronics -> integrated project

AI assistants are useful for routine repository work, but generic prompts often produce inconsistent results: broad rewrites, decorative diagrams, missing validation steps, weak engineering justification, or documentation that is not aligned with the SDR route.

Decision

The repository will maintain a lightweight AI-assisted engineering layer:

  • CONTEXT.md for shared project context;
  • .ai-skills/ for reusable engineering procedures;
  • this ADR directory for durable workflow decisions.

AI-assisted work should follow these principles:

  1. read CONTEXT.md before changing the repository;
  2. choose the closest skill from .ai-skills/;
  3. make small reviewable changes;
  4. prefer reproducible scripts, figures and checks;
  5. report validation commands and limitations;
  6. preserve the bilingual engineering nature of the course.

Consequences

Expected benefits:

  • less repeated prompting;
  • fewer forgotten quality requirements;
  • more consistent labs, figures and CI fixes;
  • better reproducibility of AI-assisted changes;
  • clearer onboarding for future contributors and AI agents.

Trade-offs:

  • the skills must be maintained when the repository structure changes;
  • overly rigid skills may need revision for unusual tasks;
  • validation still depends on actually running the relevant commands.

Initial skill set

The initial skill set covers:

  • diagnosis before patching;
  • lab creation and improvement;
  • IEEE-style figures;
  • CI repair;
  • documentation, navigation and asset verification;
  • Verilog verification;
  • DSP demo and benchmark-style outputs.