I build automations that survive production — not demos.
What I do:
— n8n workflow design & deployment (self-hosted and cloud)
— AI agents: retrieval over your data, multi-step reasoning, tool use
— Custom API integrations (REST, webhooks, GraphQL, legacy systems with no SDK)
— Data pipelines: scraping, dedup, enrichment, a schema you can actually query
What makes the difference: every flow ships with explicit error branches, retries with backoff, a fallback model where it matters, and logging. The failure mode I design against is the run that quietly produces nothing — that’s the one that costs you money three weeks later.
Recent work (details on request):
— A 7-round research pipeline that produces a 30+ page sourced report for ~$0.05/run. Primary-source facts are injected deterministically in code, because the model “forgets” them under load.
— A scoring engine with 49 regression tests and an anti-drift layer, so yesterday’s scores stay comparable to today’s.
— A lead-qualification bot: 3-step qualification, structured lead relayed to the owner in real time, 6 tracked events for attribution.
Pricing: single workflow from $400 · multi-workflow system with an AI layer $1,500–4,500 · retainer from $1,200/mo. Fixed quote before we start, free 15-min scoping call.
You own everything: flows live in your instance, credentials stay in your accounts, documentation comes with the handover. No lock-in.
Stack: n8n, Make.com, Python, FastAPI, PostgreSQL, Docker, Claude/GPT/OpenRouter, RAG & embeddings, Telegram Bot API, React, nginx/VPS.
Remote, overlapping EU and US-East hours. English C2.
Portfolio: https://nichr.tech
— Alex Kolbas