ProcessFlow

Context

Before applying an AI-first methodology to a production enterprise product, I wanted to find its limits somewhere with no stakes. ProcessFlow was that testbed: take a domain I knew well — business process discovery and improvement — and see how much of a working platform an AI-assisted workflow could actually produce.

Approach

• Built solo, using LLM assistance across specification, implementation, and iteration.
• Deliberately chose a familiar domain so I could judge output quality rather than guess at it.
• Tracked where AI assistance accelerated the work and where it quietly produced plausible-but-wrong output.

Outcome

• A working prototype, and — more usefully — a concrete sense of where LLM assistance needs human validation gates.
• Those lessons became the hallucination-mitigation practices now embedded in a production SDLC.

Stack