ShieldView

Inspiration

ShieldView was inspired by a problem every enterprise team knows too well. People share their screens all day, and one wrong window can expose a contract, customer record, API key, financial spreadsheet, patient file, or internal roadmap. We wanted to build something that protects teams in the moment, before confidential information is accidentally shown.

What it does

ShieldView acts like a privacy layer for live screen sharing. It runs in the background during calls, detects sensitive content on the screen, and blurs risky regions before they reach Zoom, Meet, or Teams. The local Mac handles screen capture and redaction, while NVIDIA Nemotron on Brev helps with deeper risk detection. NemoClaw gives teams a web interface to start protection, run briefings, and monitor ShieldView statuses, and also send back reports of the redacted screen share in case the user would like to look over it.

How we built it

We built ShieldView as a hybrid Mac and Brev system. The Mac captures the screen, performs OCR and pattern matching, applies blur effects, and sends the protected feed through a virtual camera. Brev runs the remote Nemotron service for AI analysis. We connected the two with tunnels so the local privacy pipeline and cloud services can work together.

We also integrated NemoClaw into the Brev workflow. NemoClaw runs on the Brev environment and can call ShieldView through the Mac bridge at http://127.0.0.1:8787. This lets the agent control ShieldView from a web UI, including starting privacy protection, checking health, running pre call briefings, and monitoring redaction status.

Challenges we ran into

The hardest part was making the system reliable during a live call. Screen sharing needs low latency, stable redaction, and a smooth virtual camera feed. We also had to make the Brev setup repeatable while keeping local environment files, API keys, caches, and generated data out of the uploaded project.

Another major challenge was connecting NemoClaw on Brev back to ShieldView running locally on the Mac. We solved this by exposing the Mac bridge through a reverse tunnel, so the agent could call ShieldView endpoints as if they were local to Brev.

Accomplishments that we're proud of

We are proud that ShieldView works as an end to end privacy workflow. It can detect sensitive information, blur it locally, use Nemotron for deeper analysis, and let NemoClaw control the system from Brev. We are also proud that the user experience is quiet. Once it is running, protection happens in the background.

What we learned

We learned how important hybrid architecture is for privacy tools. Local processing keeps the most sensitive screen data close to the user, while cloud AI can still add intelligence when needed. We also learned how to connect an agent running on Brev to a local Mac application through tunnels, which made the NemoClaw integration possible.

What's next for ShieldView

Next, we want to make ShieldView more enterprise ready. That means better policy controls, team level configuration, stronger audit logs, more accurate document detection, and deeper NemoClaw workflows. We also want ShieldView to support more meeting platforms and give organizations a safer default for every screen share.