Doc. AM-SPEC-001 Rev. 01 2026-09-03 Public

Alpha
Merge

A two-person software studio. We design, build and operate production systems — then keep running them.

Discipline Full-stack product engineering
Size 2 engineers
Base Israel · UTC+3
00

Summary

What this document is

AlphaMerge builds software that has to stay up. A rental marketplace serving three languages in production, a PRTG-class monitoring system with roughly 43 sensor types, an authoritative-server multiplayer game, a licensed desktop app with its own billing backend.

This page is the studio described the way we describe our own systems: numbered sections, plain claims, and a link to the code behind each one. Nothing here is a mockup — every project listed ships, or is in a named release phase.

Projects in the estate 9
Running in production 4
Primary language TypeScript
Deployment Self-operated — GCP, Docker, Caddy
Engagement Project or retained
01

Studio

How we work

We take products from nothing to running. Not just the feature work — the schema, the auth, the deployment pipeline, the TLS, the backups, the runbook for when it breaks at 2am. Two engineers who have each done every layer means there is no handoff gap and no one waiting on a specialist.

We write things down. Our monitoring system ships eighteen documents: a scope contract with explicit out-of-scope markers, architecture decision records, a phased plan with exit criteria, runbooks for backup-restore and field test. Clients inherit that, not a repo nobody can read.

We reuse deliberately. A shared baseline — Angular PWA, Express/TypeScript API, Google Identity Platform sign-in, Docker, Caddy, GitHub Actions to a GCP VM with an end-to-end health gate — was extracted from four production projects. New work starts deployed and tested on day one rather than reassembling the same plumbing a fifth time.

02

Capabilities

Six things we have shipped more than once

02.1

Multi-sided marketplaces

Listings, search, messaging, ratings, moderation, quotas and payments-adjacent flows — including binding PDF contracts with placed signature fields and a stamped audit trail.

02.2

Infrastructure monitoring

Probe-and-core architectures: schedulers, sensor catalogues, state machines with pause precedence, notification routing, remote probes and RBAC.

02.3

Realtime systems

Authoritative server rooms with per-client state filtering, WebSocket transports behind the same HTTP server, and proxies routed by upgrade header.

02.4

Search & geospatial

Elasticsearch full-text with faceted category attributes, map views with bounding-box queries, autocomplete, saved searches and alerting.

02.5

Desktop & bots

Electron apps with a strict contextBridge boundary, WhatsApp and Telegram automation, product-key licensing with a Stripe-backed issuing service.

02.6

Applied ML, privately

In-browser embeddings so prose never leaves the client, pgvector similarity on the server, and speech-to-text / translation / TTS pipelines.

03

Selected work

Product names only — client repositories are private

Rentool

Marketplace

A rental marketplace for the Israeli market — power tools, site plant, access equipment, garden machinery and event kit, rented from people nearby by the day, week or month. Hebrew, English and Arabic with full RTL, ILS pricing, location-aware search. One model carries rental, sale, service and job listings; rentals are backed by mandatory PDF contracts with e-signature and a stamped certificate.

Case study → Angular PWA · Express · MongoDB · Elasticsearch
StatusIn production
Apps3
LanguagesHE · EN · AR
SearchElasticsearch

Watchdog

Infrastructure

A PRTG-class infrastructure monitoring system in Node.js and TypeScript. Probes run checks against your devices; the core stores results, evaluates state and routes notifications. It ships as a standalone native service — systemd on Linux, WinSW on Windows — with PostgreSQL as its only always-on dependency. Roughly 43 sensor types, network discovery, remote probes and role-based access control.

Case study → TypeScript monorepo · PostgreSQL · systemd / WinSW
StatusPhase 10 — release
Sensor types~43
Packages4
DependenciesPostgreSQL only

WhatsBot Desktop

Desktop product

An Electron and Angular desktop app that auto-replies to WhatsApp and Telegram from one shared rule set, one set of menu flows, one message log and one AI configuration — only the socket differs. Sold by product key, issued and validated by an Express and Postgres licensing service with Stripe webhooks and customer accounts.

Electron · Angular · Baileys · Postgres · Stripe
StatusIn development
ChannelsWhatsApp · Telegram
LicensingProduct key

AlphaBrain

Knowledge tool

A second brain that keeps your notes as Markdown in your own Google Drive or a local folder, and draws the connections between them as a navigable 3D graph — explicit wiki-links plus implicit edges from semantic similarity. Deliberately zero-knowledge: embeddings are computed in the browser, and only the file id and a 384-dimension vector reach the server. Postgres with pgvector resolves the similarity.

transformers.js · pgvector · Socket.IO
PrivacyZero-knowledge
EmbeddingsIn-browser, 384-d
Stores3

NativeChat

Bot

Send a voice message in any language, get back what it says in yours — as text and as a voice note. No typing, no per-message language selection, no need to know what language the recording was in. Live on Telegram as @NativeChatterBot, running speech-to-text, translation and text-to-speech on GCP behind a dockerized bot.

Try it on Telegram ↗
StatusLive · Phase 1
PipelineSTT → MT → TTS
PlatformGCP · Docker

MicDrop

Karaoke platform

Digital karaoke built around the room rather than the song. Create a room, queue tracks together and sing along with friends in real time — the queue, playback state and everyone present stay in sync over WebSockets. Angular on the front, an Express and GraphQL API on MongoDB behind it, with a container-injected repository layer that keeps the socket handlers thin and stateless.

micdrop.me ↗ Angular 22 · Socket.IO · GraphQL · MongoDB
StatusLive
RealtimeSocket.IO rooms
APIGraphQL + REST
InterfaceAngular 22 · Spartan UI

Durak

Realtime game

Online and couch multiplayer Durak on an authoritative server. Colyseus rooms attach to the same HTTP server as the API, and per-client state filtering is what keeps a hand at its own seat. The board renders on PixiJS and GSAP in a lazily-loaded chunk of a zoneless Angular PWA. Every push to main deploys it.

durak.alphamerge.dev ↗
StatusIn production
RealtimeColyseus 0.17
RenderPixiJS 8 · GSAP
04

Products

What we own, as distinct from what we build for clients

Two things in the estate are ours rather than a client's, and they answer different questions.

04.1

WhatsBot Desktop

A licensed desktop product with its own billing spine — Stripe webhooks, a Postgres-backed key issuer, customer accounts and a marketing site. Telegram needs nothing but a bot token: the app holds a long poll open for as long as its window is, so there is no webhook, no public address and nothing to host.

04.2

The Boilerplate

Our factory. A baseline full-stack starter extracted from four production projects: fork it, run one rename script, and you have a deployed, HTTPS-served, tested app skeleton. It is why a new client project starts with a health-gated pipeline instead of a blank repository.

05

Engineering

The stack, as actually deployed

Frontend

  • Angular — standalone, strict, zoneless
  • React
  • PWA · service workers
  • PixiJS 8 · GSAP
  • Mapbox

Backend

  • Node.js · TypeScript
  • Express
  • Socket.IO
  • Colyseus
  • Python

Data

  • PostgreSQL · pgvector
  • MongoDB · Mongoose
  • Elasticsearch
  • Redis
  • SQLite

Platform

  • Docker — multi-stage
  • Caddy · Let's Encrypt
  • GitHub Actions → GHCR
  • GCP
  • systemd · WinSW

Identity

  • Google Identity Platform
  • JWT + rotating refresh
  • RBAC
  • reCAPTCHA v3
  • Web Push · VAPID

Desktop & ML

  • Electron — contextBridge
  • transformers.js
  • GCP Speech-to-Text
  • GCP Translation · TTS
  • Baileys · Telegram Bot API

Every project deploys the same way: GitHub Actions runs the tests, builds a multi-stage image, pushes it to GHCR, deploys to a GCP VM and holds the release behind an end-to-end health gate. TLS is automatic through Caddy. Nothing ships by hand.

06

Team

Both of us write and operate the code

Roman Michailuck

Co-founder · Engineering

Full-stack across the estate — marketplace and monitoring architecture, deployment pipelines and the shared baseline the studio builds on.

LinkedIn ↗

Vladislav Arsenyuk

Co-founder · Engineering

Full-stack across the estate — product surfaces, realtime and desktop systems, and the data layers underneath them.

LinkedIn ↗
07

Contact

Direct to both of us

Tell us what needs
to get built.

Send the problem rather than a spec — we will come back with what we would build, how long it takes and what it costs.

Studio operational Doc AM-SPEC-001 · Rev 01 Updated 2026-09-03 IL