Frontend engineer. Minsk.
I build small products end to end and the tooling I need along the way. TypeScript, React, Next.js on the front; whatever the problem needs behind it. Everything below is something I actually ship and use, not a portfolio exercise.
You paste a conversation, or drop a screenshot of one, and it writes three replies in different registers: natural, flirty, bold. The interesting engineering problem is not the generation, it is the context: a reply only works if the model reads the whole thread first, including the part where the other person went quiet. Three generations are free without an account.
On iOS it runs from the share sheet through a Shortcut, so a screenshot goes from Tinder or Hinge straight to three written replies with no upload screen in between.
Third-party trust check on ScamAdviser: English · Nederlands · Português · Español · Deutsch · Français · Italiano · Українська · Română · Русский · 简体中文 · 繁體中文 · 日本語.
Half of AuraMog is the generator, the other half is a library of guides on how texting actually goes wrong. I write them because the same questions kept coming back, and because a product that only gives you an output teaches you nothing. The full set lives on the Dating message playbooks index. A few I keep sending people:
Ghosting: what it means and how to respond
Telling an actual disappearance apart from an ordinary delay, and the one bounded check-in worth sending.
How to flirt over text without forcing it
Notice something real first, add warmth second, match the other person's participation third.
Dating profile bio: a specific, honest writing guide
Before-and-after rewrites of real bios, and why specific beats clever every time.
How to choose and answer three Hinge prompts
Picking prompts that do not repeat your own profile back at the reader.
Best first messages for Tinder, Hinge and Bumble
Eight openers, why each earns a reply, and the common ones that get left on read.
There is a reference side to it too, which turned out to be the part people actually bookmark: every Hinge prompt with an example answer, 75 lines sorted by tone, and the split-by-target collections — funny, cheesy and dirty pick up lines, each grouped for her, for him and for Tinder rather than dumped in one list.
ahrefs-mcp
MCP server over the Ahrefs API v3 — keyword research and SERP analysis from inside an AI agent instead of a browser tab. TypeScript.
vault-engine (private)
Local-first memory engine for Claude and Codex. Everything an agent has ever done, searchable, on disk, never leaving the machine. Python.
chromecast-webrtc (private)
Casting a WebRTC stream to Chromecast, receiver and sender. The protocol documentation is thin, so most of that work was reading packets.
Work, collaboration or a question about any of the above: github.com/newilu. For AuraMog specifically there is a contact page.