AI research and engineering
I research, build,and test newapproaches in AI.
Andrey Chmerev, independent AI researcher and engineer. I study local models, agent workflows and AI-assisted development, and publish the results as open tools and products.
Scroll to explore
Selected work
Research that shipped.
Speech in, text out. Nothing leaves your Mac.
Kekoso
On-device speech recognition for macOS: dictation into any app, file and YouTube transcription, call recording with speaker separation, and a local MCP server for AI agents. One-time purchase, no cloud.
Open ↗Framework for AI-Assisted Solo Development
FRAME
An open-source workflow for building products alone with Claude Code: Research → Plan → Build → Review → Ship → Reflect, with project memory, safety hooks, parallel worktrees and an unattended autopilot.
Product page →What I'm about
Research in public.
Andrey Chmerev is an independent AI researcher and engineer who builds and ships local-first AI tools. He studies local language models, autonomous agent workflows and AI-assisted software development, then publishes what works as open-source tools and products.
Two results so far: Kekoso, an on-device speech-to-text app for macOS that handles dictation, transcription and call recording without sending audio to the cloud, and THE FRAME, an open-source npm framework that lets a solo developer research, plan, build, test and ship software with Claude Code. Every shipped product is treated as the result of an experiment, and the findings are written up here, on X andReddit.
Latest writing
From the blog.
- 2026-09-16Where to list your app after launch: 134 directories rankedI submitted a Mac app to 134 startup directories and launch platforms in a week. Which are free, which pass link weight, and the cost per DR point.
- 2026-09-15Why I build alone with an AI pairNotes on running a one-person product studio with Claude Code: what works, what breaks, and why I open-sourced my workflow.