Applied AI systems. Build evidence. Operating model.

Problem

Campaign delivery breaks when intake, naming, Salesforce setup, tracking, creative handoff, DAM destinations, Workfront routing, and Pardot build instructions are scattered.

System prepared

A governed campaign intake and orchestration layer that creates campaign packages, Salesforce campaigns with returned Campaign IDs, Clarevine links, DAM image destination URLs, Workfront tickets, design briefs, and generated Pardot build instructions.

Status

Fully built and deployed with a global company, with integration APIs built. The boundary is clear: Copilot prepares the package and evidence trail; the separate Pardot build, QA, approval, and launch process remains human reviewed.

Problem

Enterprise form spam does not just create bad submissions. It pollutes Eloqua data, damages campaign reporting, wastes sales time, and creates operational noise across marketing automation workflows.

System built

Cloudflare Turnstile and Eloqua protection layer with exception queue, direct forwarding, accept/reject handling, secured reviewer access, and multi-form architecture. Accepted, deployed, and now in use by a large enterprise.

Why it matters

Proves Greg can build applied AI-assisted workflow protection around real marketing operations, with anonymised adoption evidence and an emerging Forward Deployed Engineer pattern.

Problem

High-intent prospects abandon marketing forms before becoming visible in Eloqua, Salesforce, or campaign reporting.

System designed

A governed recovery layer for partial intent capture, consent-gated recovery journeys, AI-assisted completion, and CRM handoff.

Why it matters

Designed to recover demand that already reached a form while keeping compliance and human handoff controls visible.

Problem

Audience targeting decisions made without structured evidence, scoring criteria, or human review controls at scale.

System planned

Planned AI-supported targeting workflow with scoring, evidence trails, manual accept/reject controls, queue management, export readiness, and a planned Cloudflare Worker backend with KV memory layer.

Why it matters

Planned human-in-the-loop design with evidence, scoring criteria, and operator review before any action.

Problem

Marketing teams need campaign, competitor, channel, and website intelligence pulled into one decision layer.

Architecture selected

A governed dashboard layer for turning scattered marketing signals into operator-ready insight.

Why it matters

Shows AI support for marketing strategy and evidence-led campaign decisions.

Problem

Client support workflows managed across email, files, invoices, status updates, and memory - with no shared visibility, request tracking, or billing transparency.

System built

A secure consultancy client portal for requests, retained hours, documentation, invoices, attachments, and delivery evidence. The latest production security hardening pass verified the Access/D1 trust model, tenant isolation, file handling, CSV export safety, escaping, repository hygiene, and live login path.

Why it matters

Proves Greg can build and harden the operational layer around client work. The final repo-code review found no remaining critical, high, or medium findings; only platform-layer Cloudflare checks remain tracked outside the repo.

Problem

Build progress disappears into chats, commits, and memory notes unless it is captured into reusable content.

System planned

A planned local agent that turns Tanya memory and build records into draft-only weekly content packs.

Why it matters

Converts real build evidence into reusable marketing assets while keeping human review in the loop.

Problem

Contractor-site LinkedIn content needs a controlled private scheduler, not a public social SaaS workflow.

System built

A live private scheduler for Greg Staunton personal LinkedIn publishing, covering drafts, scheduling, OAuth connection, publishing status, failure handling, and operator reporting.

Why it matters

Pressure-tests Cursor 3.0 and Cloudflare delivery on a real operational need with privacy, verification, and controlled publishing boundaries.

Problem

Local models can generate code, but reliable delivery also needs scope, checks, review, and a complete handoff.

System built

A connected DGX Spark lab: LangChain, LangGraph, Deep Agents, Mem0, Docling/MCP, tracing, and isolated checks inside the Caraval harness.

Why it matters

Reviewed local contributions have reached application integration, and the lab is being used on a significant confidential project. Broader unattended reliability remains open.

Problem

Keep time entries, timers, project access, and reports consistent as people move between tasks.

System built

A completed, enterprise-deployed time and project application with manual time, correction history, durable timers, eligible projects and tasks, and personal report filters.

Why it matters

The interesting engineering is in accurate totals, access boundaries, and recovery when the interface changes state.

Problem

Turn crawl findings and search evidence into useful actions without manufacturing metrics or amplifying false issues.

System built

An own-site crawler, stored audit history, prioritised recommendations, reports, imported search data, and a protected operator interface.

Why it matters

The build shows why trustworthy classification and visible data sources matter before adding more intelligence.

Problem

Make a large bilingual website straightforward to edit and publish without losing the relationship between source files and public URLs.

System built

Named HTML source pages, shared styles and scripts, a route map, and a dependency-free Node build that produces the deployment directory.

Why it matters

A live website becomes easier to maintain when editing, previewing, and publishing follow an explicit repeatable path.

Problem

Compress a desktop video towards a practical target file size without hand-tuning FFmpeg settings by trial and error.

System built

A Windows desktop FFmpeg compressor with probing, deterministic planning, sample pre-test, progress reporting, dependency setup, and visible startup failure handling.

Why it matters

Shows AI-assisted software delivery without pretending the product itself is machine learning. The planner is rule-based and the hard part was making the workflow reliable.

Problem

Eloqua form submissions need a durable identity stitching key where the visitor GUID is available.

System built

Page JavaScript reads the Eloqua visitor GUID where available, injects it into a hidden form field, submits it with the form, and stores it in a SYSTEM - Eloqua ID CDO.

Why it matters

Supports CDO processing, CRM matching, campaign reporting, and visitor/contact stitching without blocking submissions when GUID is unavailable.