Pavan K.

I'm Pavan Kumar. For 13 years clients called me a 10x engineer because I stuck to sound engineering practice. Today AI coding agents write most of my code, and clients see 100x output. The reason it works: I turned the practices that made me 10x into a system that validates everything an agent builds, without me becoming the bottleneck. How I build with agents: • Spec first. Every task starts as a ticket with clear acceptance criteria. Project rules live in the repo (AGENTS.md), so every agent session follows the same standards. • Tests define done. Each service has contract tests, and end to end tests follow the real user flow. All of it runs in CI on every change, so an agent cannot ship something that breaks what already works. • Small reviewable slices. One ticket at a time, subtasks as stacked PRs, each small enough to actually read. I merge only when I'm satisfied. • Regressions become tests, so the same bug never comes back. • Several agents run in parallel in isolated worktrees, so output scales while the quality gates stay the same. What I've built: • Vconic: led commercialization of a multi tenant vCon platform and scaled ingestion to 12 million calls per day on on prem H100 GPUs, with a SvelteKit analytics portal and an MCP server for AI assistants. • Comcent: founded and built an omnichannel support platform (calls, SMS, web chat, email) on FreeSWITCH, SIP and WebRTC with AI call transcription and analysis. • STROLID: integrated dealer CRMs (eLead, Tekion, DealerSocket) and telephony carriers, built an Electron dialer, added Datadog and Playwright test automation. • IETF vCon: early implementer of the standard, working with its authors. If you want software built fast that you can still trust in production, send me what you need built. Pavan Kumar