Properties: Mikhail Yakubenko
Mikhail with an X-ray of his own skull, pixelated

Mikhail Yakubenko

System
User:
Mikhail Yakubenko
Handle:
muhyrla
Age:
22 y.o.
Location:
Turkey
Occupation:
Backend developer
CPU:
Python 3.13 @ FastAPI
DescriptionBonziBuddy

I build backends in Python: Django and FastAPI, APIs, auth, integrations, bots and parsers. Comfortable with DevOps: CI/CD, containers, monoliths and microservices. Not a frontend guru, but I can vibe-code a UI when needed 😁

Installed components
  • Python
  • FastAPI
  • Django
  • Bots & parsers
  • Docker
  • CI/CD
  • Figma
  • React (vibe)
Русский
Under constructionmuhyrla.dev
Relayn - C:\Projects

Relayn

AI support for Discord communities. Answers questions 24/7 from your docs and learns from your moderators.

2025 — now · SaaS · live, with a free plan

My role: the whole technical side. Backend, Discord bot, RAG, dashboard, landing, infrastructure and deploys.

Open relayn.io

The problem

Discord communities grow faster than support teams. The same questions go round and round: how do I connect, where are the docs, why doesn't it work. Moderators burn out, and newcomers wait for hours and leave.

A bot that reads the docs

The bot joins a server via OAuth in a couple of minutes. Then you upload your docs or a GitBook: Relayn chunks them, embeds them and stores them in a vector database. It answers only from the chunks it finds and links the source, so it doesn't make things up.

Learns from moderators

When the docs don't have the answer, the bot doesn't guess: it reacts with 👀 and waits. A moderator answers, Relayn remembers it and handles that question on its own next time. All such questions land in the unanswered portal, where each takes one click to answer.

Analytics

The dashboard shows how many questions came in, how accurate the bot is (from 👍 and 👎), what people ask most and which moderators help the most. One account can run several servers, each with its own knowledge base.

Under the hood

Backend on Django and DRF with PostgreSQL, background jobs in Celery and Redis, vectors in ChromaDB, LLM and embeddings via OpenRouter, the bot on discord.py. Dashboard on React 19 and Vite, landing and blog on Next.js 16 and Sanity. Everything runs in Docker, Sentry catches the errors.