Customer story — Construction
Kajima measures jobsite work from camera footage, river levels and weather data, without being on site
Newton analyzed five years of canal site video with tide and weather data, enabling remote tracking for offsite project managers.
Summary
Kajima, which has built Japan's infrastructure since 1840, used Newton on a five-year canal-widening project in flood-prone Niigata.
- Newton read years of site video alongside weather and river-level data
- Project managers could see when equipment worked, when it sat idle, and why
- A review too large to do by hand became a search that takes seconds
Key results
Seconds
to retrieve relevant footage — a review Kajima's staff called impossible by hand
5 TB
of archived video from 27 cameras, searchable by camera, activity and date
3 streams
weather, river level and wave height matched to every day of footage
4 years
of site operations turned into daily logs and date-range summaries
The challenge
A jobsite too big to watch in real time
Kajima's canal project runs across a wide, remote estuary where river currents, wave height, and weather decide whether excavation can happen at all. Project managers answer for safety, quality, cost, and schedule every day, but they cannot stand on site all day — and construction's shrinking workforce means each manager now covers more than ever.
Cameras were already recording, but the footage wasn't usable. Nearly 12,000 videos had piled up over the project, with no practical way to find the specific hours of excavation a manager needed to check. Reviewing it by hand was, in Kajima's own words, impossible.
The stakes are high. McKinsey finds that 98% of construction megaprojects run over budget, and clients expect delays to be explained with evidence rather than estimates. Kajima set out to change how large sites are run — beginning with a goal of managing half of all jobsite work remotely.
The solution
One physical agent, every camera, plus the weather
Archetype AI and Kajima built a custom Newton Agent for the site. Newton, a physical world model, uses sensor data and camera streams to understand the construction environment: which machines are working, when work starts and stops, and how river level, wave height and weather shape each day.
The agent measured equipment utilization across excavation and towing and correlated it with environmental data brought in through Newton’s API. Managers select a camera and an activity, move to the moment on an interactive timeline, and watch the clip with detected objects highlighted — so every result can be checked against the footage itself.
Built-in feedback mechanisms enabled site staff to refine model predictions over time, improving accuracy for complex river excavation tasks through continuous usage.
Capabilities used
- Multi-camera video understanding
- Sensor fusion — video + tide + weather
- Daily log and date-range summaries
- Detected-object playback
- Expert feedback loop
How Physical AI is Revolutionizing Construction
Kajima
“It was impossible for humans to find the right video of the targeted drilling operations from 12,000 videos; Newton was able to extract these targeted videos seamlessly in seconds.”
Shinnosuke Sekihara
Mechanical and Electric Deputy Manager, Kajima
The impact
From archived footage to better planning
Newton analyzed years of stored footage after the fact and showed Kajima's managers where excavation time was lost and how weather and river conditions explained it. That record supports more realistic scheduling, backs up conversations about delays with data, and preserves the judgment of veteran staff as they retire. Monitoring live sites in real time is the next step Kajima is working toward.
Other customer stories
City of Bellevue Pedestrian Safety Agent
Trained on real data provided by the City of Bellevue, this agent enabled by Newton processes 43 GB per intersection daily, running on the edge powered by Dell's platform using inexpensive GPUs and hosted on AT&T's network.
Kajima Construction Efficiency Agent
Newton analyzed five years of canal site video with tide and weather data, enabling remote tracking for offsite project managers.
_Schedule a Demo
Put Physical AI to work
Talk with one of our Physical AI experts and learn how to deploy Newton Agents across a wide range of use cases — mapped to the outcomes you care about.
