Engineering notes

Field notes from shipped systems

Architecture decisions, failure modes, and design rationale from real robotics, AI, and backend work. Rough edges included.

Featured

May 31, 20267 min read

Building an end-to-end voice outreach pipeline in n8n: GPT-4o scripts, ElevenLabs cloning, and the tuning parameters I got wrong first

How I built an n8n workflow that turns a CRM lead into a personalized voiced voicemail in ~15 seconds — and what I learned about ElevenLabs stability, similarity boost, and why the defaults sound robotic.

  • n8n
  • ElevenLabs
  • GPT-4o
  • Automation

May 31, 20268 min read

STT → function calling → TTS: what building a voice CRM interface taught me about GPT-4o tool dispatch

How the Voice-Driven CRM Assistant works — Whisper transcription, three GPT-4o function definitions, Switch-node dispatch, Supabase memory, and the defensive edge cases that made it actually usable.

  • n8n
  • OpenAI
  • Function Calling
  • Whisper

May 18, 20265 min readLaunch readiness and idempotency

Partner fintech escrow: launch readiness is mostly about keeping state honest

Field notes from a closed-source escrow build: retries, idempotency, scheduler locks, mobile money paths, and the gap between feature-complete and launch-ready.

  • Partner Build
  • Fintech
  • Idempotency
  • Launch Readiness

May 17, 20265 min readLaunch hardening review

Sellavie.ai launch hardening: the work nobody sees until it fails

A security and launch pass across the Sellavie backend, dashboard, mobile app, CI, logging, and generated artifact hygiene.

  • Sellavie.ai
  • Security
  • CI
  • Launch Risk

May 6, 20264 min readTrial feedback and retention

Sellavie.ai feedback flow: a small feature with billing edges

Building trial feedback collection and launch incentives — what looks like a simple feature once you trace it through billing, email, dashboard state, and admin settings.

  • Sellavie.ai
  • Retention
  • Billing
  • Regression Tests

More notes