Nine AI GTM systems live, all built solo
TikTok Miner, CreativeOS and FinGuard are three of the nine below. I put each one on a public URL, and each runs an LLM pipeline with a human in the loop.
Senior Marketing Manager at ThinkingAI since 2026. Four years at PingPong before that, a 1,500-person company: Marketing Manager, Ads & Lifecycle, 2021–2023, then Senior Digital Marketing Manager, Growth & Operations, 2023–2025. And six years at ToLocal, a company of twenty-odd, 2015–2021, the last of them as VP of Marketing. Ten years of marketing before the first line of code.
When the stack can't do it, I open an editor
Most GTM teams hit a wall in the stack and file a ticket. I stopped filing tickets. After ten years running marketing for B2B fintech and SaaS, I started shipping the software myself.
At PingPong I owned marketing for the cross-border payments line end to end, the line that went from $2M to $76M in year one and past $250M by year two. Sales closed the deals and finance ran the payment rails; every campaign, channel and lifecycle touch that fed their pipeline was mine. I personally cut LinkedIn CPL from $300+ to ~$160 and Google registration cost from $90+ to ~$35, and I governed the 8-tool GTM stack that carried it all: Salesforce, HubSpot, Webflow, 6sense, Zapier, ZoomInfo, Salesloft, and GA4 with Tag Manager. Before that, as VP of Marketing at ToLocal, I ran a $10M annual media budget at 80%+ ROI leading a team of 7 in the US, with 5 more in China. ToLocal was a twenty-odd person shop in B2C affiliate performance and PingPong a 1,500-person B2B fintech.
I built the 200+ workflow layer that collapsed lead handoff from ~12 hours to 1–2 minutes. That was my build, wired into a stack I governed, feeding a sales team that worked the leads. I wrote the LLM pipeline for a messy monthly report that used to take one to two hours by hand. It now runs in under five minutes and turns one cleaned dataset into eight reports. My LinkedIn title line reads “GTM Systems Builder” because that is the job I do.
Nine live URLs
I ordered these by what they settle. The systems are the artifact, each on a URL you can open right now. The case studies show the same method applied inside a company, with other people in the loop. In the writing you can watch me reason before I knew the result.
Systems, all live
Competitor RadarI split it on cost: a cheap model labels every competitor ad, and a frontier model reads the strategy off the labels. TechSpyIt runs a real browser and captures the HAR. It states confidence and admits blocked scans. CreativeOSThe generator reads policy as structured records while writing, then scores against them. Built independently, no customers yet. FinGuardCompliance keeps the regulator's rulebook as data and maintains every rule in it. Career CapybaraOne console governs thirty-five prompts, each pinned to a model with an audit trail. All nine, side by sideI list all nine with the stack each one runs on.The same method, inside a company
Reusable multi-agent systemsAgent teams with guardrails. The review layer caught a failure before a customer saw it. Agent-native reportingIt compiles the weekly status from source systems, and other people depend on it. Content enginesAgents produce and package the work, and I sign off before anything goes out. AI sales intelligenceI turned a dead roster into a ranked, reachable book sales worked. AI video pipelineBrief to finished film, no crew. Same argument, different medium.How I think, written down
Search ads research as an agent pipelineFour models, each doing what it does well on 3,156 keywords. I explain the routing. Automating a messy monthly reportWhere the accuracy broke, and why the fix belonged in the review step. An AI lead scoring assistantScoring one question at a time so a rep can see the reason beside the score. A scored content pipelineI turned an editorial judgment into a rubric a model applies consistently. Automation for a one-person support deskThe smallest version of the whole idea. I state the limits.From a marketing pain
to a public URL
01 · Start from a pain I've paid for
Every system began as a GTM problem I lived with as a marketer: ad copy dying in policy review, leads going cold over a ~12-hour handoff, a monthly report eating an afternoon. I write the problem down with a number attached before I write any code.
I ask for the number the pain costs before I build02 · Ship the smallest system that closes the gap
CreativeOS is the pattern. I built it to write platform-ready ad copy for Google, Meta, LinkedIn and Bing with formats and character limits enforced, checking every line against platform ad policy and custom compliance rules as it writes. Next.js, an LLM API, Vercel. On an independent build I am the only reviewer, which is why it ships fast and where its proof stops.
Ship the narrow thing, then stop adding to it03 · Wire it into the stack
A tool that lives outside the stack is a demo. At PingPong I built the workflow layer myself: 200+ workflows across an 8-tool stack I governed, which took lead handoff from ~12 hours to 1–2 minutes. The nine systems are separate builds, each on its own public URL. Software that touches revenue has to sit inside the plumbing, so I build the plumbing too.
A tool nobody has to open is a tool nobody uses04 · Keep a human in the loop
Every one of the nine has a human approval point by design, because GTM software touches money and brand. The model drafts, extracts or flags; a person approves. I measured what happens without that gate as a marketer, so I refuse to ship without it as an engineer.
Automation stops where accountability startsDoes “marketer who codes” mean scripts?
One TikTok URL in, a queryable intelligence system out
No. TikTok Miner is a full ETL and RAG system: paste a TikTok URL and it transcribes the video, extracts frame-level metadata, embeds everything into a Pinecone vector store, and serves RAG chat plus engagement analytics on top. React 19 and TypeScript on the front, multiple AI models behind it, built solo.
The messy monthly report I automated went from one to two hours of hand-assembly to under five minutes, and it writes eight reports off one cleaned dataset. The problem was smaller and I worked it the same way: I measured the pain, then shipped the pipeline to a URL anyone can visit.
9 systems live · full ETL + RAG · report time 1–2h → under 5 minDaniel is a fantastic member of the team with his ability to blend very deep technical knowledge with great interpersonal skills. He is able to quickly understand business requirements and then present a solution in a motivational way in terms appropriate for any level of the audience.
Emma YuSenior Product Manager, TikTok
Hiring for AI GTM?
Hand me the GTM problem. I will ship the system that fixes it and give you the URL.