What I do.

I built the shared Claude Code workspace the content team writes in, for 16 client brands. It's a private Git repository everyone opens in Claude Code, with a folder for each brand holding its context and voice notes, and skills for the jobs the team does every week, like drafting posts or turning a meeting recording into notes. Opening a session pulls the latest version and closing it pushes, so a change one person makes reaches everyone within seconds.

Drafting starts by finding a few of the client's own published posts to use as examples, and then Claude writes the draft and checks its own work. That retrieval step is what my internship research tested, and the write-up has the numbers.

I ran the team's onboarding in July and wrote the setup guides, and the non-technical team now runs it day to day.

The work was my M1 internship from May to August 2026. Now it's a contract to the end of the year.

I designed ACE-NSI's method for assessing Shadow AI in the regulated firms it advises. That's the industry's name for staff pasting work into AI tools their company never approved. The method lists the AI tools in use and rates each one on how sensitive the data going into it is and whether any control covers it, which together give a risk level. The findings are then mapped to GDPR and the EU AI Act, with NIS2 and DORA added where they apply. I wrote it up as a sample report on a made-up company, so the deliverable can be shown without anyone's real data in it.

I also tested two AI-security platforms hands-on. One caught a test string typed into ChatGPT and warned the user in real time. The other inventoried the agents and MCP servers in a demo environment, skills included, and graded the risk of each, with findings tagged to the OWASP Top 10 for LLM applications. They cover different parts of the problem, and I wrote a guide to which control fits which situation.

People often ask whether Microsoft Purview already covers this, so I checked Microsoft's documentation. Purview blocks prompts for about 18 AI apps, in Edge only, and Claude and Le Chat aren't among them.

I've also worked on their lead generation, and built a Python pipeline on the Claude API for researching target companies and keeping the CRM in step with the outreach.

I also set up Claude Code for people directly: small businesses, and an individual sales leader who wanted it built around their day-to-day work.

It usually starts the same way. Claude Code reads the files with the person there while they explain how the work actually runs, and we agree a structure before anything moves. They end up with one working folder that Claude Code reads every time it opens. In it is a description of the work in their own words, plus skills or templates for the jobs they repeat, like reviewing a deal or planning a campaign.

For the sales leader the skills draft and never send, and everything runs on their own machine. I design the setup and write the prompts, and they run them, so I never see their files. After that it's coaching, on calls where they do the work and I help.