What I’ve built and measured
Alma, a voice agent that interviews talent
Alma is the voice agent at Shakers that interviews and evaluates freelance talent. We built it there, and I led the product as AI Product Lead until October 2026.
- 43% to 89%
- of interviews finished
- Over 85%
- less cost per interview
What was going on
An automated interview is only worth anything if the candidate finishes it.
What we built
I led the product and the technical work on Alma. Most of the work was the conversation itself, with turn taking, silences, interruptions and calls that drop.
The CRM at Magnific (formerly Freepik)
I was Tech Lead of the CRM at Magnific (formerly Freepik), a platform with millions of users. The CRM handled tens of millions of contacts.
What was going on
Sales, support, billing and marketing worked with separate tools and manual handoffs.
What we built
I started on my own, building the Salesforce foundation from scratch and the integration and automation layer around it, with Braze, Chargebee, APIs and internal workflows. Then I put together a team and led it.
90%
less response time in customer service
Design and development of the AI implementation strategy at Shakers
What was going on
Giving everyone access to AI leaves each person to sort out context, tools and workflows alone.
What we built
We built an internal marketplace of features on top of Claude, with skills, MCP connectors and agent workflows that any team could install and reuse.
How I evaluate usage
I separate two things almost everyone mixes up. Setup is what each person has set up. It can go from assisted tools to having their data connected and their own skills, or to orchestrating several agents.
Usage is how they use it. When reactive, they ask for something and accept what comes out. When deliberate, they direct it, review against a criterion and take responsibility for the result. When rigorous, they define the criteria first, put controls in place and, when something fails, fix the system and not just the answer.
| Assisted | Extended | Orchestrated | |
|---|---|---|---|
| Rigorous | |||
| Deliberate | |||
| Reactive |
> 90%
of AI adoption