The main projects, with the starting problem, what I built and the result. Details are in the linked notes.
AI in production
An AI agent working on a live WordPress site
- Problem: optimizing a page took ten minutes of copy-paste between the admin and an AI assistant, with as many chances to make a mistake.
- Built: a WordPress mu-plugin giving an agent limited access (pages allowed one by one, a backup before every write, a dry run), and written procedures to optimize a post or rebuild a page.
- Result: the agent reads the live page, queries Search Console, writes once I approve and checks the result. No more copy-pasting HTML.
- Notes: AI agent and WordPress · skills · rules learned
Marketing and sales automation
A tooled-up prospecting chain
- Problem: incomplete contacts, poorly personalized messages and sales tasks getting lost.
- Built: n8n workflows that enrich contacts with Apollo, compute personalization variables with an LLM, feed Lemlist and remind sales reps of their tasks in Teams every day.
- Notes: Apollo enrichment · grammar variable · Teams reminders · LinkedIn posts
Business signal detection
- Problem: sales databases tell you a company exists, not why to contact it now.
- Built: a system combining scraping, technology detection, job postings and competitive monitoring, then ranking signals with an LLM to prepare the sales team’s work.
- Notes: overview · lead machine · technology detector
Web, SEO and measurement
dolist.com redesign
- Problem: four websites for one company, an offer hard to read and a scattered SEO history.
- Built: a single site, with a new architecture, rewritten content, a redirect plan and new tracking. Run with an agency; I owned architecture, SEO, content and acceptance testing.
- Note: Redesigning dolist.com from A to Z
Lab
Personal projects to test the same methods on different ground.
- Plex collections from SensCritique rankings, scripts freely available
- Cleaning up a Shield TV Pro and a Shield TV tube, memory measured before and after