Most AI systems fail not because
they don't work — but because
nobody uses them.
Enterprise sales reps were spending the majority of their time manually assembling client proposals and course outlines from fragmented course content — a process that was slow, inconsistent, and impossible to scale. No tooling existed to support this.
I designed and built a full-stack content intelligence platform: a structured GitHub repository as the canonical content source, a React/Vite frontend that reads from GitHub and collects enterprise client parameters, and a Claude API integration layer governed by the atomic modules logic. The system reduced manual workload by 80%, allowing to increase the client flow fivefold.
"I have waited half a year to see something like this."
— CEO, live product review session
The client — an investigative podcast brand with a published archive of 17+ episodes — was producing high-value interview content that disappeared into static audio files. No searchable knowledge base. No content reuse. No compounding value. Each episode existed in isolation. Significant editorial intelligence was being generated and then lost.
I designed and built a three-tool automation stack (Whisper AI → Cursor → GitHub) and a four-type knowledge atom model to make every interview compound. Each transcript is processed into structured atoms: Voices (direct source quotes), Patterns (recurring themes across guests), Tensions (contradictions worth investigating), and Leads (story threads for future episodes). The GitHub repository functions as a living editorial brain — queryable, versioned, and directly connected to downstream distribution.