Applied AI systems for enterprise marketing workflows.

Tanya Build Cockpit

Problem: AI-assisted build sessions drift without structure, memory, or consistent QA checkpoints.

AI-assisted build orchestration layer using structured prompts, project memory, QA gates, implementation packets, and agent role separation. The operational scaffolding behind every pl8ypus build.

Shows the build discipline and AI collaboration layer that makes the other systems production-shaped rather than experimental.

Translation AI

Problem: Marketing translation done manually, inconsistently, with no protected-term enforcement or review trail.

Advanced governed translation build for regulated marketing assets. Operator workflow, pilot evidence, protected terminology, deterministic backend contracts, QA gates, review status, and human approval before output is treated as final.

Proves repeatable AI workflow design with control layer and evidence capture - not just a prompt.

Campaign Copilot

Problem: Campaign delivery breaks when intake, naming, creative handoff, tracking, and Pardot build instructions are scattered.

Governed campaign intake and orchestration for enterprise marketing operations: structured naming, Salesforce campaign creation, returned Campaign ID, Clarevine tracking, DAM destination URLs, Workfront tickets, design briefs, generated build instructions, and handoff into a separate human-reviewed Pardot build workflow. Deployed with a global company.

Marketing Form Protection

Problem: Enterprise form spam pollutes Eloqua data, damages campaign reporting, and wastes sales time across marketing automation workflows.

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

Enterprise-used workflow. Greg's second fully built AI-assisted solution now used in a large organisation, and a practical step in the Forward Deployed Engineer journey.

Lead Recovery AI

Problem: High-intent visitors abandon marketing forms before the CRM ever sees them.

A concept for governed abandoned-form recovery in enterprise marketing: partial intent capture, consent checks, email and SMS recovery, AI sales-agent completion support, and clean Eloqua/Salesforce handoff.

Audience Finder AI

Problem: Audience targeting decisions made without structured evidence, scoring criteria, or review controls.

Planned AI-supported audience and targeting workflow with scoring, evidence trails, and manual accept/reject controls. The concept keeps queue management, export readiness, and memory as future implementation goals.

Planned as a human-in-the-loop design with clear evidence and review controls.

Marketing Intelligence AI

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

Selected architecture for analysing LinkedIn activity, search signals, website behaviour, campaign performance, and cross-channel opportunities. Designed as a boardroom-style intelligence layer for marketing teams that need evidence before campaign decisions.

Shows how AI can support marketing strategy, not just production tasks, by turning scattered signals into operator-ready insight.

Client Portal

Problem: Client support workflows live across email, files, invoices, status updates, and memory.

A portal sounds simple until it has to behave like a real support system. Requests, updates, documents, billing visibility, attachments, admin controls, audit trails, identity, and multiple users attached to the same client account.

That is where the toy version breaks. This product now runs as a Cloudflare-native client workspace with Access authentication, D1 records, R2 file storage, tenant isolation, admin controls, and production hardening.

Proves Greg can build the operational layer around client work, not just isolated prototypes. This is the kind of system that supports embedded enterprise problem-solving and production-minded delivery.

Weekly Build Agent

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

A planned local agent that reads Tanya memory and build session records, then produces draft-only weekly outputs for blog posts, email updates, and LinkedIn posts. Human review stays in the loop. No auto-publishing, no external AI calls, and no live system changes.

Turns real build evidence into reusable marketing assets while keeping Greg in control of what gets published.

Social Scheduler AI Build

Problem: Professional LinkedIn content needs a controlled private publishing workflow.

A private scheduler for Greg's personal LinkedIn profile, covering drafts, scheduling, OAuth connection, publishing status, and failure handling.

AI Lab

Problem: Turn local model output into software that can be tested, reviewed, and handed over.

The story of the NVIDIA DGX Spark lab and Caraval: bounded local workers, isolated checks, useful delivered tools, and the failed attempts that improved the workflow.

Recorded delivery evidence and human acceptance, with broader unattended reliability still to prove.

Hourloom

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

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

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

SEO Command Center

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

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

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

Segunda Escolha

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

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

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

TargetSize

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

A Windows desktop FFmpeg compressor with probing, deterministic target-size planning, sample pre-test, progress reporting, setup scripts, and visible startup failure handling. AI helped build the software, but the compression planner is rule-based.

Shows the practical engineering loop: plan, build, test, debug a silent startup failure, and keep the final product claims honest.

Eloqua Visitor Identity Layer

Problem: Eloqua form submissions often arrive without a durable visitor/contact stitching key.

A page-level identity capture pattern that 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 for downstream matching and reporting.