Work / what we build and operate

Systems that run at 3am, while you sleep.

We don't do case studies with logos and glowing quotes. We do systems — live, in production, handling real volume. Below is what we build. If you want evidence the numbers are real, the live dashboard is pulled from production this second.

Client work is under NDA — we don't name names, we don't show logos. What we can tell you is what we've built, the stack behind it, the scale it runs at, and the problems it solved. Combined with the live dashboard, that's harder evidence than any testimonial page.

/ 01 · Workflow Platforms

High-availability workflow platform

A production customer-facing workflow platform with state management, SLA timers, document automation and full audit trail.

What we solved

  • Multi-step customer journeys with branching logic
  • Case-state machines, SLA timers, deadline escalation
  • Document generation + e-signature integration
  • Integrations with telephony and third-party APIs
  • Full audit trail suitable for regulated environments
  • GDPR-by-default data handling

Stack

WordPressNode.jsPHP 8.3 MariaDBRedisnginx Docker

Custom WordPress + bespoke Node backends on hardened Docker containers; shared-infra isolation; off-site encrypted backups.

/ 02 · Infrastructure

Hardened multi-tenant production hosting

Multiple live production sites on a single hardened box, each fully isolated, with per-site WAF and rate limiting. Zero cross-tenant incidents.

What we solved

  • Per-site Docker container isolation (read-only FS, CapDrop ALL)
  • Independent nginx vhosts with per-site WAF rules
  • Scoped fail2ban jails per application
  • Default-server catch-all against Host-header confusion
  • Automated TLS renewal (Let's Encrypt) for every vhost
  • Per-site nightly backups with retention policy

Stack

nginxDockerUFW fail2banLet's Encryptsystemd Ubuntu 24.04

Host nginx handles TLS termination + reverse proxy; every application runs in its own locked-down container. We deliberately run without Cloudflare — the site you're on is the proof.

See live status
/ 03 · AI & LLM

Production Claude-powered customer assistant

A live Claude Haiku 4.5 chatbot with knowledge-base retrieval, Anthropic prompt caching, 11-layer abuse control, lead extraction, and a hard daily spend cap.

What we solved

  • Knowledge base cached on Anthropic side (90% input discount)
  • Per-IP hourly + daily limits, session message caps
  • Origin validation (strict URL parsing, not prefix match)
  • Bot-UA blocking before the model is called
  • 4-tier graceful degradation on spend cap
  • Email auto-detection → lead capture + notification

Stack

Claude Haiku 4.5Anthropic API Node.jsPrompt caching Dockernginx

This site runs it. Click the chat button bottom-right and you'll talk to the exact system described above.

See chats + cache hit rate live
/ 04 · Voice & Telephony

Dialler + call-centre infrastructure

Asterisk + ViciDial dialler stacks tuned for audio quality, least-cost routing, and uptime.

What we solved

  • Asterisk PBX configuration + dial-plan tuning
  • ViciDial predictive-dialler setup + campaign management
  • SIP carrier integration + least-cost routing
  • AMD (answering-machine detection) calibration
  • Call recording, conference bridges, supervisor monitoring
  • Agent provisioning, SIP extension lifecycle, WebRTC
  • Designed to scale to high-volume campaigns

Stack

AsteriskViciDialSIP WebRTCMariaDBLinux

Dialler operations at volume is a rare specialism. Most engineering studios won't touch it. It's where we do some of our best work.

Got a system you need to actually work?

If your platform is fragile, your compliance is hanging by a thread, your dialler drops calls, or your AI integration is a hack pretending to be production — we can help. No pitch deck. We'd rather show you what we've already shipped, then scope yours.

Start a project