Work that survived production.

Runs customer applications on isolated Docker infrastructure with managed SSL, domains and snapshots — provisioned in minutes, not days.

Problem

Small teams need production-grade apps — n8n, Supabase, WordPress — without hiring anyone to manage servers. Managed platforms are either too narrow or too expensive.

Constraints

Per-customer isolation. Provisioning had to be fully automated — no manual steps between checkout and a running app. Cost low enough for solo founders.

What was built

Docker-per-customer provisioning with an application catalog (n8n, Supabase, WordPress, static hosting). Automated SSL issuance, domain wiring and snapshot scheduling. A control panel that hides every server concept.

Reliability safeguards

Isolated containers per customer. Automated snapshots with restore. Provisioning rollback on failed launches. Per-instance resource limits so one tenant cannot starve another.

Business outcome

Instance counts and uptime figures are being pulled from provisioning records and monitoring, and will be published once verified. No estimates.

Keeps lead follow-up running around the clock for real-estate and mortgage operators — capture, CRM and automated nurture in one system.

Problem

Agents and loan officers lose deals to slow follow-up. Leads arrive from ads, portals and referrals into different inboxes, and nobody owns the response.

Constraints

Compliance-safe outbound messaging. US-market timing. Existing tools had to keep working — the system wraps them, it does not replace them.

What was built

Lead capture funnels, CRM automation and follow-up sequences with reporting, concentrated on real-estate and mortgage operators in Tulsa, Oklahoma, serving US and UK markets.

Reliability safeguards

Human approval on outbound sequences. Delivery monitoring. Opt-out handling built into every flow.

Business outcome

Uptime and CRM figures will be published after verification against monitoring and the production database. No estimates.

Answers service-business calls 24/7, books appointments, and hands complex calls to a human.

Problem

A missed call is a lost job. Service businesses miss calls during work hours and lose every after-hours call to voicemail.

Constraints

Real phone lines, real customers. The agent had to fail gracefully — a confused caller must always reach a person or a reliable callback, never a dead end.

What was built

A telephony-to-LLM pipeline with a booking flow wired into the business calendar, orchestrated through n8n. The agent qualifies the caller, books the appointment, and logs the call.

Reliability safeguards

Human-handoff path on low confidence. Full call logging. Escalation rules per call type. Fallback to voicemail plus automatic callback if any component fails.

Business outcome

Call volumes will be published from execution logs after verification. Capability facts stand on their own: 24/7 answering, booking flow, human handoff.

Plans, produces, schedules and publishes finished social content. You approve, it ships.

Problem

Consistent social content takes a team: strategist, writer, designer, scheduler. Most founders have none of those roles filled.

Constraints

Output had to be finished — reels, carousels, captions — not drafts. Nothing publishes without explicit approval.

What was built

A content pipeline that plans strategy, produces finished assets, schedules them and auto-publishes across channels, with an approval gate in front of every post.

Reliability safeguards

Approval gate before publishing. Per-channel rate and window rules. Full audit trail of what shipped where.

Business outcome

Generated-content and early-access account counts will be published from the production database after verification. No estimates.

Routes content jobs across OpenAI and Anthropic models in production, with a review gate on every step.

Problem

Different models are good at different jobs, and every provider has bad days. Single-provider pipelines stall when the provider does.

Constraints

Provider-agnostic by design. Any step must be reviewable before its output moves downstream.

What was built

A multi-model writing and orchestration system running OpenAI and Anthropic models side by side: routing by task type, prompt pipelines, retry and cross-provider fallback, structured output review.

Reliability safeguards

Cross-provider fallback. Retries with budget caps. Human review gates on final output. Full run logging.

Business outcome

Generation volumes will be published after verification. The system runs production workloads across the CTL stack daily.