Founder
Döppelscript
Döppelscript is an AI writing tool I founded in early 2026, built around a single premise: generic AI writing has a 'tell' that readers can spot from a mile away, and the only way to defeat it is to learn each user's voice from their existing writing and generate new content in that specific voice. The product targets professionals who want to be active on LinkedIn without their posts reading like ChatGPT output.
I shipped a working MVP in three days on Next.js 15, Express/TypeScript, Postgres on Neon, Cloudflare R2, BullMQ workers, and the Anthropic Claude API. The architecture mirrors the patterns I established across the rest of my fullstack AI portfolio, with async voice-extraction jobs running in BullMQ workers and generation routed through a quality-judge loop that scores every draft against the user's voice signature.
The voice extraction pipeline analyzes uploaded corpora (LinkedIn exports, blog posts, PDFs) to derive a durable voice signature covering sentence rhythm, vocabulary preferences, rhetorical patterns, and banned terms, then reuses it on every generation. Combined with hard anti-slop rules (no em dashes, banned-word enforcement), the output stays indistinguishable from the user's own writing.
Key Achievements
- Founded Döppelscript, an AI writing tool that learns a user's voice from their existing writing and generates LinkedIn-ready content in that voice, shipping a working MVP in three days on Next.js 15, Express/TypeScript, Postgres (Neon), Cloudflare R2, BullMQ workers, and the Anthropic Claude API.
- Engineered a per-user voice extraction pipeline that analyzes uploaded corpora (LinkedIn exports, blog posts, PDFs) to derive a durable voice signature (sentence rhythm, vocabulary preferences, rhetorical patterns, banned terms), then reuses it on every generation to eliminate the 'AI tell' that defeats generic writing tools.
- Built an in-line quality judge and refine/rewrite loop that scores every generation against the user's voice signature and flags drift toward generic AI cadence, paired with hard anti-slop rules (no em dashes, banned-word enforcement) to keep output indistinguishable from the user's own writing.