Engineering lab

Lab

Smaller tools, working methods and the systems I intend to build next. Each entry states its status and the evidence behind it.

Tools and methods

Things I built to make the main systems work: infrastructure control, how I work with AI agents, and this site.

Status: Prototype

Observe-first autoscaling for Zone 32-19

Railway Environment Controller

A small TypeScript service that reads Railway metrics every minute and decides whether services should scale, but by default only records what it would do.

  • Reads CPU, memory, replica count and deployment status through the Railway GraphQL API.
  • Scales on hysteresis with sustained windows, per-service and global cooldowns, min and max replicas and a budget cap.
  • Refuses to replica-scale stateful services such as databases, Redis, volumes and singleton workers.
  • Writes one JSON audit line per service per run, including decisions it blocked or skipped.
  • Applying a change needs an explicit mode, dry-run off and, for production, an approval file. 61 tests cover policy, config, locks and scaling.
  • TypeScript
  • Node.js
  • Railway

Evidence: Zone-32-19 monorepo, Zone-32-19-Environment-Controller (v0.3.0)

Status: Active development

How I build with AI coding agents

Agent-governed engineering

Most of my recent work is written together with AI coding agents. What makes that work is governance: written repository rules, decision logs, review agents and tests that hold the boundaries.

  • Every major repository carries its own operating rules for agents, and architecture decisions are recorded as ADRs or decision logs.
  • Zone 32-19 has custom review agents for routes, databases, migrations and security, plus checklists for schema changes.
  • J.A.R.V.I.S. is developed by several agents on parallel branches that I integrate, and it exposes project context to agents over MCP, read-only except for five tools that record derived state.
  • I keep authorship and integration: architecture, boundaries, review and merges are mine.
  • Claude
  • Model Context Protocol
  • GitHub Actions
  • Git

Evidence: Agent rule files across repositories, the Zone 32-19 review-agent definitions, and the J.A.R.V.I.S. branch history and decision log

Status: Active development

NIKOLAY.RUNTIME, version three

This portfolio

A static Astro and TypeScript site. Content is typed data, architecture diagrams are generated from that data, and nothing ships a client framework.

  • Content collections with schemas for every case study, and a technology registry based on evidence rather than self-rating.
  • A diagram engine that renders typed nodes and edges to SVG, with an accessible list version for phones and screen readers.
  • A strict Content Security Policy with hashed scripts, self-hosted fonts, no trackers and no cookies.
  • Astro
  • TypeScript
  • HTML and CSS

Evidence: This repository

Next systems

Planned work, deliberately outside my current stack. Nothing here is built yet, and nothing here is presented as if it were.

Status: Planned

Banking and ledger platform in Java, Spring Boot and Angular (working name)

LedgerForge

My next major system, and a deliberate step outside TypeScript: a financial platform where correctness is the product.

  • Accounts, balances and a double-entry ledger with an immutable transaction history.
  • Transfers and payment flows with idempotency, clear transaction boundaries and reconciliation.
  • Audit logging, roles and permissions, risk signals, reporting and an administration console in Angular.
  • A modular monolith first, with strong test coverage and observability, rather than microservices for their own sake.
  • Java
  • Spring Boot
  • Angular
  • PostgreSQL

Evidence: Planned. No repository exists yet.

Status: Concept

Block-based fantasy sandbox

Untitled voxel world

A 3D open world built from blocks: survival and building in a medieval-fantasy setting with cities, creatures and dragons, designed from the start to be extended by mods.

  • Engineering topics to solve: chunked world streaming, procedural terrain, persistence, an entity-component system, networking and a plugin API.
  • Inspired by the block-building sandbox genre, aiming for a more polished and extensible world.

Evidence: Concept. No code yet.

Paused

Long-term ideas on hold.

Status: Paused

Realistic large-scale fantasy game

Velnethyrn

A long-term idea for a realistic, large-scale 3D fantasy world. It is paused, and the repository holds only a placeholder.

Evidence: Placeholder repository only.