Cobalt is an AI system that measures how work happens in physical operations from camera video. It shows where the time goes, what to change, what each change is worth, and whether the change held.

What kinds of operations does Cobalt work with?

Any operation where the work is physical and can be filmed: plant floors, fabrication shops, warehouses, installation sites and repair bays. Cobalt builds a video model for each process, so it does not depend on a pre-built model for that industry.

What if our process is unusual?

Most are. Cobalt is configured for each process: which activities to tell apart, what counts as value-added work, and which questions matter. The configuration changes as the process does, so a new product or a new step does not mean starting over.

How is Cobalt different from hiring a consultant?

A consultant sees a few days of the operation and leaves a report. Cobalt measures every shift, answers follow-up questions from footage already recorded, and keeps tracking after changes are made.

How is Cobalt different from computer vision for PPE or quality?

Those systems are built for one narrow task. Cobalt builds video understanding for a specific process, so it can measure time use and flow across the whole process and answer new questions about it.

How much does Cobalt cost?

Cobalt starts with a fixed-fee diagnostic, priced from the five figures depending on the number of processes covered. Ongoing measurement is a monthly subscription.

Does Cobalt replace the operation's engineers?

No. It gives whoever runs improvement, whether an engineer, an operations manager or a lead, measures they would otherwise have to collect by hand.

About Cobalt

Cobalt is an AI system for physical process transformation. It measures how work happens in physical operations, from camera video and the business's existing data, and shows where the time goes, what to change, what each change is worth, and whether the change held.

San Francisco · Founded 2026

What Cobalt does

Process improvement

Cobalt baselines the floor from video, traces lost time to its root cause, proposes ranked changes, quantifies each one, and tracks the result.

Process documentation

A value stream map of how a unit moves through the plant, and illustrated SOPs for the work at each station. Both are produced from the same video.

Plant layout simulation

A 3D model of the floor, used to compare layout options and their expected throughput before equipment is moved.

Safety monitoring

The same video is reviewed for unsafe conditions. Observations are reported by area and shift, with the clip behind each one.

How Cobalt compares

Businesses get eyes on a physical process in one of three ways today: an expert on site, single-purpose computer vision, or Cobalt.

Expert on site

A consultant, an engineer or a vendor team running a time study

Sees the whole process: Yes
Answers new questions as they come up: Yes
Runs continuously: No, only while present
Handles unusual processes: Yes
Cost: High, per visit

Single-purpose computer vision

PPE detection, part inspection or people counting

Sees the whole process: No, one narrow task
Answers new questions as they come up: No
Runs continuously: Yes
Handles unusual processes: No, only the tasks it ships with
Cost: High, per use case

Cobalt

Cameras and video AI built for the specific process

Sees the whole process: Yes
Answers new questions as they come up: Yes
Runs continuously: Yes
Handles unusual processes: Yes, configured for each one
Cost: Low, per process

What makes Cobalt different

It covers the whole process, not one task: Single-purpose computer vision answers the one question it was built for, such as whether a hard hat is worn. Cobalt classifies all the activity in every covered area, so time use, flow and the constraint come from the same footage.
It answers questions nobody planned for: When a new question comes up, such as where the crew is losing time or what changed on Tuesday, Cobalt answers it from footage already recorded. Single-purpose computer vision would need a new model built first.
It keeps measuring after the first report: A time study covers a few hundred moments over a few days, and stops when the visit ends. Cobalt measures every shift in every covered area and keeps going, so the business can see whether a change held.
It is configured for each process: Most operations have work no off-the-shelf model was built for: a custom fixture, a one-off part, a step done differently on second shift. Cobalt is set up for each process, including which activities to tell apart, what counts as value-added work and which questions to answer, so unusual and variable work is measured as well as standard work.
It does not need an engineering team: A business does not need industrial engineers of its own to get value from Cobalt. It needs a walk-through at the start and a few minutes each week from a supervisor or lead.

Who Cobalt is for

Any business with complex physical operations. The buyer is usually the operations leader responsible for the process, or the private equity operating partner responsible for improving a portfolio company.

Manufacturing and fabrication
Machining, welding, assembly and finishing, in job shops and high-mix plants.
Equipment installation and field service
Crews that install, commission and service equipment on site.
Logistics, warehousing and distribution
Receiving, picking, packing and loading.
Food and beverage processing
Batching, processing, packing and sanitation.
Maintenance, repair and overhaul
Repair bays and overhaul lines for vehicles, aircraft and heavy equipment.
Construction and prefabrication
Offsite prefabrication, and repeated work on job sites.

The team behind Cobalt

Cobalt brings together three backgrounds: Charlie's long-standing work in financial simulation, Emiliano's research in visual understanding, and Nelson's work in operations and transformation consulting. The founders share an interest in real-world problems with a measurable benefit to American industry.

Large companies keep operational excellence teams on staff. The hundreds of thousands of mid-sized businesses with physical operations mostly cannot. Cobalt was started to give them that expertise continuously, through AI.

Nelson Arnous
Co-founder and CEO
Nelson spent six years in operations consulting at BCG and Silver Lake, focused on major transformations at some of the largest companies in the world. At Silver Lake he built the firm's process diagnostic tooling, used with more than 20,000 employees across its portfolio companies. He then led product at Compound, a tech-enabled wealth advisory firm.
Emiliano Garcia-Lopez
Co-founder and CTO
Emiliano researches video understanding and spatial reasoning. He has published at COLM, with further work under review at NeurIPS, and that research is the basis of Cobalt's video pipeline. He is a Felicis AI fellow and studied at Carnegie Mellon.
Charlie Davidmann
Co-founder and COO
Charlie ran enterprise sales at Pipe, where he grew the origination program from $1M to $100M a month, and has built and sold production AI systems since 2024. He has built financial simulations for most of his life, is a former banker, and has competed in skeet shooting for the Great Britain national team.

How Cobalt works

The full sequence

Setup: Under a week, on existing cameras or cameras Cobalt installs. Operations are not interrupted.
First results: About two weeks after kickoff.
After that: The core measures update continually, new changes are suggested as the work changes, and each change is tracked from adoption to impact.
What the business provides: A walk-through at the start, access during setup, and a few minutes a week from someone who knows the work.

Key facts

Company name: Cobalt
Type: AI system for physical process transformation
Founded: 2026
Founders: Nelson Arnous (CEO), Emiliano Garcia-Lopez (CTO), Charlie Davidmann (COO)
Headquarters: San Francisco
Website: cobalt-research.com
Core offering: Continuous measurement and improvement of physical processes from camera video
Capabilities: Process improvement, process documentation, plant layout simulation, safety monitoring
Industries: Manufacturing, equipment installation and field service, logistics and warehousing, food and beverage processing, maintenance and repair, construction
Setup: Under a week, on existing cameras or Cobalt's
First results: About two weeks from kickoff
Pricing: A fixed-fee diagnostic from the five figures, then a monthly subscription
Data ownership: The customer owns its video and everything derived from it
Worker privacy: Measures are reported per station or area, never per named worker
Alternatives: On-site time studies and consultants, single-purpose computer vision
Contact: contact@cobalt-research.com

About Cobalt

What Cobalt does

Process improvement

Process documentation

Plant layout simulation

Safety monitoring

How Cobalt compares

Expert on site

Single-purpose computer vision

Cobalt

What makes Cobalt different

Who Cobalt is for

Manufacturing and fabrication

Equipment installation and field service

Logistics, warehousing and distribution

Food and beverage processing

Maintenance, repair and overhaul

Construction and prefabrication

The team behind Cobalt

Nelson Arnous

Emiliano Garcia-Lopez

Charlie Davidmann

How Cobalt works

Key facts

Frequently asked questions