Selected Work | Marklar Systems

@Marklar: dev.marklar.systems

# project lifecycle metadata; no live service telemetry

Hub-and-spoke distributed system connecting VPS, Hal, and Vindaloop via SSH tunnels and MCP protocol. Taskwarrior as distributed journal, Telegram for human-in-the-loop orchestration.
description
A pragmatic distributed system built from primitives: SSH tunnels for connectivity, MCP (Model Context Protocol) for structured RPC, Taskwarrior as a distributed event log, and text files for state. No Kafka, no Kubernetes — just boring technology that composes predictably. The agent itself is the orchestrator, calling MCP tools across nodes to manage infrastructure, security monitoring, and deployments.
stack
SSH
MCP
Taskwarrior
Python
Claude Code
tags
distributed-systems
ssh
mcp
infrastructure
writing
./posts/How I Stopped Overengineering and Learned to Love SSH
source
https://dev.marklar.systems/root/marklar.systems-frontend
Full-stack e-commerce platform migrated from WooCommerce to a custom Vue/Nuxt + NestJS + Prisma stack. Monorepo with pnpm workspaces, REST API, and automated sync pipelines.
description
A complete e-commerce platform rebuild, replacing a legacy WooCommerce setup with a modern full-stack architecture. The monorepo houses a Nuxt 3 frontend, NestJS API with Prisma ORM, and Python sync tooling for data migration and ETL. Features product catalogs, category management, order processing, user auth, and automated sync from external data sources. Deployed with Docker Compose and GitLab CI/CD.
stack
Nuxt 3
NestJS
Prisma
Python
PostgreSQL
Docker
tags
ecommerce
vue
nestjs
monorepo
prisma
writing
./posts/Elecrecycle When Excel Is Your Source of Truth
source
https://dev.marklar.systems/root/elecrecycle-vue
AI-powered Zendesk support assistant using local LLMs (Qwen3 via Ollama) with RAG on business knowledge bases. Domain-driven design with test-validated outputs.
description
A Zendesk ticket automation system that uses local large language models for retrieval-augmented generation on business FAQ knowledge bases. Built with NestJS + LangGraph.js, running Qwen3:14b on an RTX 5070 GPU. Features include business hours scheduling, GPU resource state machine (coordinating between LLM inference and other workloads), 270-320s per ticket processing, and a comprehensive QA test corpus with validation guardrails against hallucinations.
stack
NestJS
LangGraph.js
Ollama
Qwen3
TypeScript
Docker
tags
llm
zendesk
rag
nestjs
ai
writing
./posts/SAV LLM Building a Zendesk AI Assistant
source
https://dev.marklar.systems/root/sav-llm-back
Tor-aware terminal observability tool that turns outbound IPv4 flows and local Tor circuit metadata into an interactive Braille world map.
description
A privacy-conscious network visualization tool built for the terminal. Torbit captures outbound IPv4 traffic with Scapy, enriches destinations through local-first GeoIP, and renders routes, counters, and recent flows with Rich. Optional Tor Control Protocol integration maps local circuits through authenticated SAFECOOKIE sessions, while reverse DNS, remote GeoIP, and active traceroute remain explicit opt-ins.
stack
Python
Scapy
Rich
MaxMind GeoIP
Tor Control Protocol
Pytest
tags
network-observability
tor
privacy
tui
python
writing
./posts/Making Tor Traffic Visible Without Weakening Privacy
source
https://dev.marklar.systems/root/torbit
Real-time EVE Online intelligence TUI combining seeded systems, killmail activity, directional scans, and explainable opportunity scoring.
description
A Go terminal application that turns fragmented EVE Online wormhole intelligence into a ranked decision surface. It resolves corporations and alliances through ESI, retrieves rate-limited killmail data from zKillboard, parses pasted directional scans, and composes independent scoring rules into an inspectable result. Ports and adapters keep the domain independent from Bubble Tea, HTTP clients, and JSON persistence.
stack
Go
Bubble Tea
Lip Gloss
ESI API
zKillboard
JSON
tags
go
tui
eve-online
threat-scoring
ddd
writing
./posts/Turning EVE Online Intel Into a Decision Surface
source
https://dev.marklar.systems/root/snitchboard
Defensive exposure monitor that discovers leaked API credentials in authorized public GitHub scopes, validates findings, and stores only one-way fingerprints and masked fragments.
description
A NestJS security service for authorized monitoring of credential exposure across public GitHub repositories. Quick and deep search strategies feed a bounded worker pool; provider-aware patterns, entropy checks, contextual filters, and endpoint validation reduce false positives. PostgreSQL retains a SHA-256 fingerprint, masked fragments, repository context, and remediation metadata without persisting the raw credential.
stack
NestJS
TypeScript
PostgreSQL
Octokit
TypeORM
Blessed
tags
security
secret-scanning
github
ddd
worker-pool
writing
./posts/Finding Leaked API Keys Before They Become Incidents
source
https://dev.marklar.systems/root/gitmagnet
Multi-spectrum satellite imagery analysis toolkit. Downloads Sentinel-2 data, computes spectral indices (NDWI, NDVI, NBR), generates interactive maps and before/after comparisons.
description
A Python toolkit for processing Sentinel-2 satellite imagery. Automates download, band stacking, cloud masking, and spectral index computation for environmental monitoring. Supports multiple analysis profiles: hydrology (NDWI), vegetation (NDVI), burn severity (NBR), and urban change detection. Generates interactive HTML maps, time-series charts, and correlation matrices. Applied to conflict damage assessment in Mariupol.
stack
Python
Sentinel-2
GDAL
NumPy
Folium
tags
python
satellite
gis
remote-sensing
writing
./posts/Terra Lens Because Europe Launches Satellites and Nobody Uses Them
source
https://dev.marklar.systems/root/terra-lens