Most of my time right now goes into AI: getting it into people's hands safely, and building the infrastructure underneath it.
in progress
A walled garden for AI developers
We have AI developers building tools for our portfolio companies. They need real development environments with package managers, model APIs, and room to experiment. Each portfolio company's data has to stay with that company, and none of it can leak into the rest of the firm. So I'm designing a separate environment for them from the ground up:
- identity
- Their own identities and access groups, so permissions follow the project, not the person's day job.
- network
- A segmented network for development traffic, kept apart from corporate systems.
- vpn
- Dedicated VPN profiles with their own rule sets, the same approach as the travel laptops.
- zscaler
- Separate Zscaler profiles tuned for developers, not for the average office laptop.
- certificates
- Custom certificate bundles so Node, Python, and Rust tooling trusts our TLS inspection instead of failing on it.
- jamf
- Custom Jamf scripts that set up each machine the same way, every time.
The certificate piece is the kind of thing that eats a week if nobody plans for it. TLS inspection swaps in its own certificate, and npm, pip, and cargo each keep their own idea of which certificates to trust, so installs start failing with errors that don't explain themselves. The fix is a bundled CA file and a script that points every toolchain at it:
# roughly what the Jamf script sets up
export NODE_EXTRA_CA_CERTS=/opt/certs/bundle.pem # node / npm
export REQUESTS_CA_BUNDLE=/opt/certs/bundle.pem # python requests
export PIP_CERT=/opt/certs/bundle.pem # pip
export CARGO_HTTP_CAINFO=/opt/certs/bundle.pem # rust / cargo
entra id · meraki · anyconnect · zscaler · jamf · node · python · rust
-
in progress
AI across the whole firm
I'm helping roll out AI tools company-wide. That covers which tools are approved, how sign-in and access tie into our identity provider, and how people learn what data is and isn't okay to use with them.
-
in progress
AI agents in containers
I'm packaging AI agents into containers so each one gets only the network access, credentials, and files it needs, and can be torn down and rebuilt cleanly. When something goes wrong, the blast radius is one container.
-
in progress
Security tooling for AI
I'm evaluating and deploying tools that watch how AI is used: which apps people connect to, what data goes into them, and what agents are allowed to do on their own.
-
in progress
A business evaluation framework on Azure
I'm building a framework on Azure that uses AI to help evaluate businesses. It pulls in company data, runs it through models, and produces a structured assessment the investment team can work from. I'm handling the infrastructure, identity, and security around it as well as the tool itself.