Andrii T.

I build AI automation that removes repetitive work from teams — and I build the software around it. On my last project a review that took a person 40-50 minutes now takes 2-4 minutes of human approval: the work still gets checked by a human, they just stop doing it by hand. Over the last 5 months I designed and shipped the production system behind that, and it now handles a marketing team's daily review workload end to end. What that system does: it watches a Jira board, pulls creative assets from Dropbox, runs them through ffmpeg/Whisper and Claude against a written brand ruleset, posts a structured verdict back into Jira and Slack, and waits for a one-click human approval. It has a kill switch, per-zone pause controls, regression tests that run before any rule change goes live, error alerting, and audit logging. It has been running in production on real tasks with zero failed executions. I am not a no-code prompt person. I am an engineer who ships production software, so the automation I build has the things automations usually lack: guard rails, idempotency, dead-letter alerts, rollback, and a human approval gate wherever a mistake would be expensive. The engineering side: • iOS & macOS — Swift, SwiftUI, AppKit. I own the macOS and iOS clients of Spellar (spellar.ai), an AI meeting assistant: audio capture, on-device transcription with WhisperKit, speaker diarization, menu-bar architecture, App Store / Setapp / direct release flavors. • Web & backend — React/TypeScript front end, Scala backend, MongoDB, REST APIs. I take contract changes across all layers at once instead of throwing them over the wall. • Chrome extension — built a Manifest V3 recorder extension (tabCapture, offscreen documents, service workers) end to end, from the capture pipeline to the upload contract with the backend. • AI plumbing — Claude and OpenAI GPT APIs, prompt design under safety constraints, structured output, retrieval from real business data, and multi-step pipelines that must not silently drop work. How I work: I ask what the actual bottleneck is before I write anything, I show you the working thing early, and I tell you plainly what is proven versus what is still assumed. If a task is better solved without me writing code, I will say so. Typical engagements: • Automate a repetitive internal process (intake → decision → notification) with a human approval step • Connect systems that do not talk to each other — Jira, Slack, Gmail, Google Workspace, Dropbox, Notion, a database • Add an AI review or classification step to a workflow a person currently does by hand • Build or take over an iOS/macOS app, or a full-stack web feature Tools I use daily: n8n, Claude API, OpenAI GPT API, PostgreSQL, Slack API (slash commands, Block Kit, modals), Jira REST API, Google Docs/Drive/Gmail APIs, Dropbox API, webhooks with HMAC signature verification, Docker, Swift/SwiftUI, TypeScript/React, Scala, Python. If you have a process someone on your team runs by hand every day, tell me what it is and I will tell you honestly whether it is worth automating.