← Cambrian Weekly Blog
Sentinel · Weekly Update

Project Sentinel Weekly · June 22–26 (Week 1)

Aditya NagachandraAditya Nagachandra Friday, June 26, 2026 ·5 min read

Sentinel is Cambrian’s site management platform — one platform that, per jobsite, surfaces live and recorded camera streams, 3D scans, analytics, and , all behind a unified google auth based access system for employees and external parties.

This week we went from zero to a working v1: the authorization spine is live on GCP, live streamed site footage plays through gated signed HLS, and the dashboard is on a public test host.

We will push to make this stable on ios device and make it as technologically friendly as possible (ie : Face-ID based login) for easy use.

Sentinel dashboard — Sites view with live cameras at 2360 Kelton Avenue.
Current state of platform
Sentinel wiring topology — ingest, edge hub, cloud stack, and clients.
Current Platform Topology

What got built, last week to this week

Frontend. Auth-gated dashboard with live HLS cameras (iOS-safe).

Backend. Grant-gated APIs for streams and scan storage (Firestore + signed GCS).

Hosting. Moved to a persistent GCP VM for deployment and hosting.

Time Lapses

Outdoor and indoor framing

Sped-up edge recordings from 2360 Kelton — outdoor site progress and indoor/front-of-house views from the same recorder pipeline that feeds gated live HLS.

Wednesday, June 25, 2026

Outdoor — pit cam, rear yard

Thursday, June 26, 2026

Indoor / front-of-house wide angle

Camera views

Still frames from the Kelton site cameras — the same feeds that power live HLS in the dashboard.

2360 Kelton Avenue — camera 1, pit cam (rear yard / excavation).
2360 Kelton Avenue — camera 1, pit cam (rear yard / excavation).
2360 Kelton Avenue — camera 2, front-of-house wide angle.
2360 Kelton Avenue — camera 2, front-of-house wide angle.

How can I try Sentinel?

Open Sentinel and sign in with the prefilled test admin account. Pick 2360 Kelton Avenue → Cameras for live HLS, or Scans to view the point cloud.

Goal early next week:

  1. Pick two scans → Align → Show diff
  2. Bonus: How can we add a metric to these scans to signify the ‘progress’ for a certain stage (eg: framing)
  3. Jackpot: Add a ‘valuation/payout’ function for each ‘diff’ created

What’s the goal for next week?

  1. Video analysis and analytics platform v1 and integration
  2. Worker utilization metrics from site footage
  3. Task-related progress tracking (per trade / stage); OSHA compliance (future goal)
  4. VLM-based video analytics — benchmark against Memories.ai
  5. Tidy up Scans tab — point-cloud viewer, align two scans, show diffs
  6. Finalize schema and point cloud registration

Why we need repeatable, high-quality scans

The value of Sentinel’s scan pipeline is in comparing one day to the next — not in any single point cloud on its own. Registration and diff only hold up when captures are made the same way: same device, same Record3D settings, similar lighting. Walk the site in a consistent path (exterior first, then rooms), keep good coverage of walls and structure, and close the loop back where you started. When the routine slips — a different phone, a rushed walk-through, or gaps in overlap — alignment gets noisier and it’s harder to see real progress from one day to the next.