Shazil
Mohammed

A Queensland HVAC contractor needed a sales pipeline that ran without a sales team: daily lead flow, qualification, outreach, and booked callbacks.

Full multi-agent operations platform. Daily AEST-scheduled scraping of Google Maps, Airtasker, hipages, and council DA feeds. AI-driven BANT lead scoring, automated email and SMS outreach, Google Calendar-confirmed callback booking. Five-page CRM: dashboard, leads, kanban, bookings, and a heat pump unit catalogue covering Carrier, Toshiba, Daikin, and Mitsubishi.

A small brand studio based in Auckland needed to implement AI into their workflow to increase brand pilot output (more concepts, faster iterations) without losing the senior judgement that justifies premium pricing.

Nine-section operating strategy and a seven-stage AI workflow: brand intake, brand brain, content generation, calendar, approval gate, distribution, optimisation. Three productised service tiers, mapped to the studio's existing client base.

Senior hydraulics engineers spend hours cross-referencing standards and running calculations. An agent could surface that work in minutes if accuracy and traceability were built in correctly.

Orchestrator routing to four specialist sub-agents: design calculations, compliance, field diagnostics, documentation. Three-block system prompt strategy: persona/scope, chain-of-thought with explicit assumptions, structured output enforcing summary + calculations + standards citations + risk flags. RAG over actual AS/NZS 3500 documents.

Build a consumer identity product that feels personal, mystical, and shareable, without an app store cycle or a real backend.

Gamified self-discovery app combining MBTI, astrology, numerology, and love languages into a unified Soul Profile. Pixel-art avatars, 8 characters × 4 emotion states with evolution unlocks, AI-generated narrative reports, PDF download, profile comparison, 2D avatar chatroom, all in a single HTML file (~7,700 lines).

Multi-platform setup tasks require both browser interaction and terminal commands. Normally that's two separate human passes.

Paired prompt files that hand off cleanly between two agents. Cowork drives the browser: creates accounts, builds automations, gathers credentials. Claude Code injects credentials, deploys via Vercel CLI, runs validation tests, writes a launch summary. One human kick-off. Two agents finish the work.

Teams hard-code one LLM into everything and overpay, sending trivial requests to frontier models and quality-critical requests to cheap ones.

A drop-in API gateway that routes every request to the right model. Four routing strategies (cheapest, fastest, best quality, balanced) plus a policy engine for per-team rules and SHA-256 response caching so repeated prompts cost zero. Full dashboard: spend, latency, cache hit rate, per-model breakdown.

Construction inspections produce hundreds of site photos that a human must manually cross-reference against NCC 2022 and Australian Standards. Slow, inconsistent, expensive.

Upload → analyse → review → map → report. Claude Vision reads site photos and flags defects; each finding is mapped to the specific NCC 2022 / AS clause it breaches, scored by severity and confidence, then queued for a human inspector to accept or reject before a branded PDF report is generated.

Every business says "we should be using AI" yet has no structured way to turn its departments, workflows, and governance requirements into an operating system an AI can actually run.

CAIOS is a Chief AI Officer command centre. A guided intake wizard maps the business: departments, workflows, knowledge sources, risk levels, approval gates. From that model it generates a complete, deployable AI operating system: root and per-department CLAUDE.md memory files, agent definitions, a prompt library, a governance pack (AI policy, human-approval matrix, data-privacy rules, incident-response playbook), and a phased implementation roadmap, exported as a ready-to-run file architecture, with an AI-readiness score tracking coverage.

Build a self-sustaining digital product business converting cold social traffic into paid customers with no team and no ad spend.

Five-page animated quiz generating personalised result profiles via streaming Anthropic API calls. System prompt caching and compact input formatting bring cost-per-completion under one cent. 29-page PDF product, three-email ConvertKit automation, Gumroad delivery, full security middleware and rate limiting on Vercel.