I don't use AI tools.
I build them.

Seven SEO tools, built from their spreadsheet

Their entire methodology was a master checklist — 1,374 points across 7 modules, run by hand, per client, every time. I turned each module into its own tool. Six are built and verified against live client domains; only the local/GBP one is still queued. Together they do what SurferSEO, Clearscope and NeuronWriter do — plus the part none of them measure: whether the answer engines actually cite you.

• Six of seven builtA senior SEO's judgment, written down once and runnable by anyone on the team. The three scored modules alone carry 683 points across 214 checkpoints.
• Earned over verifiableChecks with no data leave the denominator instead of being guessed into it. The score never claims more than it can prove.
• Every fix is pricedFindings sort by point value, each with the evidence that produced it and what to do about it.

A Google Ads agent with write access — and a gate it can't cross

It reads live account state through the API, fans out to five specialist auditors in parallel, and drafts every change it wants to make: negatives, keyword pauses, budget moves. Then it stops at zero changes applied and waits for a human. Every mutation is logged and reversible.

• Five parallel auditorsTracking health, search terms, buyer intent, keyword waste and RSA assets — each with its own checklist.
• Intent filteringBuilt for a US collections client where the same keywords are searched by CFOs and by debtors. Getting that wrong fills the pipeline with nothing.
• 96% less execution timePart of a Claude Code + MCP suite that monitors 7 ad accounts in real time, flags issues and pushes campaign changes. Manual reporting is gone entirely.

Six content workflows. This one ships a blog end to end.

AI blogs are cheap. Blogs that get cited by AI Overview, ChatGPT and Perplexity are rare — and the difference is entirely in the work before anyone writes. So I built a workflow per content type. The blog pipeline below runs five phases with a human gate at phase four, then an 18-rule gate across four sections — pre-writing, sourcing, writing and a final quality pass — before a single file is written out.

• A person picks, the system ranks96 signals in from Reddit, Hacker News, Trends and Search Console, scored against live GSC positions. The tool never decides the topic.
• 18 rules, four sectionsPre-writing, sourcing, writing, quality. The strict ones: process content comes from internal docs, never internet research — and every stat gets traced or dropped.
• Five assets, not one draftSEO brief, article, QA report, social pack and image brief — all from the same run.

Five dashboards. This one tracks ABM down to who clicked.

Reporting used to be a weekly export nobody could ask a follow-up question of. So I built the tracking layer: FastAPI, SQLite, straight off the platform APIs, deduped on write — backend, frontend and data layer, all mine. The one below tracks a LinkedIn thought-leadership campaign against a fixed 893-company target list, and answers the only question the CEO actually asks: which of our accounts are warming up, and who inside them is clicking?

• 355 of 893 engaged60 clicked, 295 warm. A 5.91% click rate against a sub-1% B2B benchmark — roughly 6×.
• Titles, not just logosCEO, Co-Founder and Founder are the top three clicking. That's the proof the targeting landed on buyers, not on job seekers.
• It refuses to flatter youOn-list companies only. Warm means repeat impressions, never a click. Nothing off-target is allowed to inflate the number.

I'm a designer first. The agent just executes at volume.

You ask for “a creative eye for visual storytelling and brand design.” The eye can't be prompted. Mine comes from freelancing posters and key art in the Telugu film industry — where a weak layout gets seen by a few million people — and from being the de facto design lead at Revv Growth.

So I did what I did with SEO and content: wrote my own judgment down as a pack and pointed the agent at it. Five execution skills on a design system on a reference bank of creative that converted. This page was built with it.

• The eye is the input, not the outputFilm work is unforgiving about composition, type and hierarchy. That's the standard the pack encodes — not a model's house style.
• Never starts from a blank promptEvery request routes through a reference bank first, then a direction, then execution. Anti-generic by construction: no purple gradients, no centred everything, no Inter.
• Spec-checked on the way outAssets validated against platform dimensions and safe zones, with a manifest listing what was produced and what's missing.