All markets

Tech / Software

I build AI-native systems
that still make sense in production.

Multi-agent orchestration, real infrastructure, durable state, risk controls, and ops discipline — not prompt theatre. Everything below is running or has shipped.

19MCP tools live
4Domains on VPS
97Tests passing
8+Docker containers

AI Infrastructure

A real multi-agent operating layer — not a wrapper around a chat API.

Claude, Codex, Gemini, and local Ollama models all operate as interchangeable workers over shared durable state on a dedicated volume. No single model is a point of failure. The rule is simple: if it is not written to the drive, it did not happen.

  • Custom MCP gateway server with 19 production tools: shell_exec, ssh_exec, sqlite_query, wp_rest, hostinger_dns, stalwart mail control, openai_workers with failover, finance_query
  • Three additional custom MCP servers: WordPress (18 tools), videogen (Sora/Veo/ElevenLabs), bamclone (RAG + WhatsApp bridge)
  • DPAPI-encrypted credentials vault — 20+ production secrets, vault CLI for humans, vault_lookup.py for automation-safe retrieval
  • All tool calls logged to PostgreSQL ai_core.tool_calls. New barriers appended to _meta/barrier_playbook.jsonl. New tools registered in _meta/registry.csv.
  • Multi-AI coordination contract codified: agents read canonical state, do not re-brief from scratch, hand off via structured packets
See case study →
Multi-agent control plane architecture: MCP gateway with 19 tools, PostgreSQL logging, and DPAPI credential vault

Production E-commerce Architecture

A V2 gateway dark-deployed alongside V1 — no customer risk, real shadow traffic.

FocusGoods V2 is a complete second-generation multi-tenant fulfilment gateway. It receives mirrored Stripe webhook traffic, verifies signatures, queues via BullMQ/Redis, and holds at v1_weight=100 / v2_weight=0 behind a hard human gate.

  • Multi-tenant PostgreSQL with RLS: transaction-scoped SET LOCAL app.tenant_id, verified non-leaking after commit, zero unscoped reads
  • Async webhook pipeline: Fastify → Stripe sig verification → BullMQ → per-tenant token-bucket fair-share scheduler. Interactive lane exempt from throttle.
  • Rights screening engine: Levenshtein (0.80), Soundex phonetic (0.85), pHash visual similarity with Hamming distance. 18/18 tests passing.
  • Quality gate: structural validation → Gemini 2.5 Flash LLM comparator → three-tier verdict. Fail-safe: API error escalates, never silently passes. 21/21 tests passing.
  • Canary traffic controller: 10% incremental ramp, consecutive failure detection, atomic rollback with audit trail. 20/20 tests passing.
See case study →
T-V2 gateway architecture: Stripe webhook ingestion through BullMQ, RLS database, and automated canary controller

VPS & Production Web Infrastructure

Four live domains. Eight Docker containers. One person.

A production VPS running a full stack: Traefik reverse proxy, PostgreSQL 16, Redis, Stalwart mail server (DKIM Ed25519 + RSA, SPF, DMARC, ManageSieve), PM2 Node.js apps, and a WordPress fleet — all managed via MCP tooling from a local terminal session.

  • Stalwart v0.16.5: Ed25519 + RSA-2048 DKIM, programmatic Sieve script deployment via Python (port 4190), documented API limits and exact workarounds
  • FocusGoods.co.uk: Node.js/Express, SQLite, Stripe Checkout, Printify fulfilment, transactional email pipeline (6 templates), cron daily drop emailer at 08:30, admin retry endpoints
  • WordPress fleet of 7 sites — Dockerised, behind Traefik, managed via WordPress MCP server with 18 tools covering SSH, REST API, WP-CLI, FTP, Elementor, SEO, cPanel
  • Hostinger DNS fully automated via API: A, MX, SPF, DMARC, DKIM, CNAME — all scriptable with documented format quirks
  • Grafana monitoring on BQM telemetry stack: network metrics, hardware metrics, latency/jitter/loss dashboards
Discuss infrastructure work →
VPS production stack with Traefik, 8 Docker containers, 4 live domains, and Stalwart mail server

Fit assessment

Where I'm most useful

Strong fit

AI-native system designProduction backend opsMCP / tool gateway developmentMulti-tenant architectureAsync queue systemsTrust-preserving automationVPS / container operationsCredential vault design

Solid

Full-stack Node.js / ExpressStripe integrationPython / FastAPIPostgreSQL schema designWordPress at scaleSEO technical structureEmail deliverability

Adjacent / learning

TypeScript (improving)React / Next.js (familiar)Kubernetes (concept-level)Rust (interested)

Looking for an engineer who ships and operates?

Available for senior engineering, AI infrastructure, or technical operations roles and contracts.