PDE / IPDE — Garth Puckerin

NL→

Spec to Code

HITL

Human-in-the-loop

Any

LLM Provider

The Vision

PDE (Prompt-Driven Engineering) / IPDE (Intelligent Prompt-Driven Engineering) is an AI-as-infrastructure development environment. The core idea: natural-language specifications should be first-class inputs to a software project — not just prompts in a chat window, but governed artefacts that produce reproducible, inspectable outputs.

Give PDE a requirement in plain English and it produces a scaffolded repository, tests, CI configuration, and a review-ready PR — with human-in-the-loop checkpoints at each stage so engineers stay in control of what gets committed.

Architecture

Input Layer
Natural-language specification (VS Code / Electron UI)
                    ↓
Spec Parser
Requirement decomposition  ·  Constraint extraction
Dependency analysis        ·  OpenAPI schema generation
                    ↓
Provider-Agnostic LLM Layer
GPT  ·  Claude  ·  Gemini  ·  Local models
                    ↓
Generation Pipeline
Repo scaffold  →  Implementation  →  Tests  →  CI config
                    ↓
Quality Validation Loop
Static analysis  ·  Test execution  ·  Consistency checks
Human review checkpoint  →  Approve / Redirect
                    ↓
Output
Review-ready PR  ·  Audit trail  ·  Spec linkage
          

Key Design Decisions

Provider-agnostic adapters: The LLM layer is abstracted behind a unified interface — swap GPT for Claude, Gemini, or a local model by changing one config line. No prompt rewrites required.
Reproducible runs: Every generation run is deterministic given the same spec and seed. Runs are logged with full input/output tracing, making it possible to audit exactly what the model was asked and what it produced.
Guardrailed output: Generated code is validated against static analysis rules, test suites, and schema constraints before a human is asked to review — filtering out obvious quality failures before they reach an engineer's inbox.
Spec linkage: Every generated file maintains a traceable link back to the original natural-language requirement. When requirements change, PDE can surface which files are potentially affected.

Connection to L&D Work

PDE / IPDE directly extends skills I developed in LMS administration — specifically the focus on governed, auditable workflows and ensuring that complex systems can be operated by teams rather than individuals. The same principles that make a well-configured LMS reliable (clear roles, documented processes, audit trails) apply to AI-assisted development.