01
One system of record
Keep sensor data, labels, schemas, approvals, and exports connected to the dataset version that produced them.
Company
Independent infrastructure for Physical AI.
Avala connects sensor data from robots, vehicles, and drones to human + AI review, versioned dataset releases, and the model feedback that starts the next cycle.
One connected loop
TraceableSensor data
MCAP · LiDAR · video
Dataset releases
Versions · lineage · permissions
Model feedback
Evaluate · drift · re-label
Every model outcome can lead back to the data.
Founder story
Avala began with the data loop behind Tesla Autopilot.
Before founding Avala, Emal Alwis joined Tesla's Autopilot AI team as its third software engineer and reported directly to Elon Musk. He worked on systems that turned fleet edge cases into model improvements at production scale.
That experience shaped Avala: a connected loop that preserves sensor context, review decisions, dataset lineage, and model feedback. We build the data foundation so Physical AI teams can focus on their machines and models.
Emal Alwis
Founder & CEO
At a glance
Avala, in five facts.
- Category
- Independent infrastructure for Physical AI.
- What we run
- The loop that turns a fleet's own sensor data into its next model: 4D reconstruction, human + AI labeling, quality control, versioned releases, and the training handoff. Labeling is one stage of that loop.
- What customers own
- The data, the labels, and the model. Customers train in their own pipeline; Avala runs the infrastructure that makes each release better than the last.
- Founded
- 2021, San Francisco. Founder and CEO Emal Alwis was the third software engineer on Tesla's Autopilot AI team.
- Backed by
- Wonder Ventures, MaC Venture Capital, Flybridge Capital, and Valor Equity Partners.
Why Avala
Physical AI improves when its data has memory.
The bytes multiply. The hard part is knowing which data mattered, what changed, who approved it, and which release reached training. Avala keeps that operating history intact. Every deployment produces the exceptions its models have not seen; keeping that history intact is what makes the next release better than the last.
Operating proof
Built in production, not in theory.
Our teams run real sensor-data workflows inside Avala. That operating depth becomes reusable software—quality gates, dataset versions, and feedback loops—not a trail of one-off deliveries.
Where we are going
Owners of their intelligence, not tenants of someone else’s.
By 2036, Avala is the open infrastructure on which any operator of machines in the physical world turns its own experience into its own intelligence, with data, models and deployment under its own control, so that everyone can be an owner of their intelligence, not a tenant of someone else’s. It begins with the loop we run today: make one workflow repeatable, carry its history across more models and teams, and help every fleet learn from its own experience.
Read our visionGlobal operations
Expert judgment, connected to the system.
A trained network across five continents handles the cases models cannot, while every decision stays tied to the dataset release.
Operating network
5 continentsAvala
North America
South America
Africa
Europe
Asia
Human decisions stay inside the same governed workflow as automation.
How we build
Principles for durable infrastructure.
The platform should make the right operating behavior easier: one record, visible evidence, and human judgment where it changes the outcome.
02
Evidence at every gate
Make review decisions, quality signals, and release history visible so teams can reproduce what reached training.
03
People where judgment matters
Use automation for throughput. Route ambiguous, safety-critical, and domain-specific cases to trained reviewers who keep context across releases.
Long-term partners
Backed by partners who share our mission
Build with us
Build the infrastructure Physical AI will depend on.
Join a team working across data systems, model operations, and people-first execution for robots, vehicles, and other machines that learn from the physical world.
Start with one workflow