I build the systems that move pipeline: CRM architecture, lifecycle automation, attribution, and production AI agents.
Growth operations, not growth hacking. Everything I build is documented, reports its own failures, and has a human review gate where judgment matters.
emrekent.com · LinkedIn · FlowKit Systems
A CRM and revenue engine that survives staff turnover. 3 connected pipelines, 28 documented workflow modules, 15+ client sub-accounts, covering lead capture through offboarding. The systems support 170 invoices and approximately $319K in recorded collected billing.
Two production AI agents, with the boring parts built in. Productized 52 strategy reports, roughly 312 strategist hours, into 2 agents with observability, version control and failure tracking. Report generation fell from about 6 hours to 9 minutes 24 seconds at $2.21 compute. Directional comparison, and human review still applies.
Server-side attribution that does not lie to you. A multi-source acquisition bridge feeding web and paid leads into CRM through 5 Meta Conversions API workflows, logging 2,175+ contact enrollments.
Paid acquisition from zero. $5,700+ across Apple Search Ads, TikTok and Reddit driving 2,642 installs at a $2.19 blended CPI, improving CPI 14% while scaling spend.
| Principle | What it means in the repo |
|---|---|
| Failure is visible | Retries, telemetry and explicit failure states. A system that never reports failure is not being measured |
| Human review is a node | Not a disclaimer in the footer. It is a gate the flow must pass through |
| Documented as built | Not written afterwards. The handover is part of the build |
| Scoped down and said so | A narrow version with its limitation stated beats a broad one that overpromises |
TypeScript · Next.js · React · Python · Supabase (PostgreSQL) · n8n · GoHighLevel · Vercel · GitHub Actions · Zod · Meta Conversions API · GA4
Building in public at Emre Automates. Commercial work through FlowKit Systems.