@deepenai

Safety, productivity, velocity!

 Santa Clara, CA
deepen.ai
Joined January 2017
  • Tweets192
  • Following187
  • Followers364
36Media
Safety Pool™ brings together leading automotive industry organizations, policymakers, regulators and stakeholders to build a shared collaborative future for Safe Autonomous Driving and ADAS systems. Excited to have @Aptiv on board! #SafetyPool #AV #Safety #ADAS @wef
2
1
8
Deepen AI retweeted
NVIDIA DRIVE
@NVIDIADRIVEOct 23
Replying to @MohMusa @nvidia
Congrats, @MohMusa! Making safety measurable is a mission worth building toward. Great to have @deepenai in the Halos SAFE ecosystem. 🙌”
1
1
117
Deepen AI retweeted
NVIDIA DRIVE
@NVIDIADRIVEOct 23
Replying to @deepenai
Congrats on the PASAP launch, @deepenai! Excited to help make AV safety more measurable together. 🤝
2
2
130
Introducing PASAP, Deepen AI’s Physical AI Safety Assessment Program. We’re bringing data quality, Safety Pool™ scenarios and measurable safety evidence to AV evaluation with @NVIDIADRIVE Halos SAFE. deepen.ai/press-releases/dee…

Deepen AI Introduces Physical AI Safety Assessment Program (PASAP) to Support AV Evaluation with...

PASAP combines data characterization and scenario-based benchmarks to help developers expand test coverage and generate traceable safety evidence.

deepen.ai
1
2
185
We are excited to launch our Physical AI Safety Assessment Program in close collaboration with @nvidia @NVIDIADRIVE Halso SAFE
NVIDIA DRIVE
@NVIDIADRIVEOct 21
At #NVIDIAGTC Berlin, European mobility leaders showcased how NVIDIA Hyperion supports safe autonomous driving and robotaxi fleets. We also introduced Halos SAFE to accelerate AV safety validation, and Alpamayo 2 Edge for in-vehicle reasoning. Read more: nvda.ws/4z0pstH
31
A team spends six months and a few hundred thousand dollars collecting robot demonstrations. Another team already had that data, sitting on a drive. The buyer couldn't find it. The owner had no safe way to sell it. Refinery is built to fix both.
A team spends six months and a few hundred thousand dollars collecting robot demonstrations. Another team already had that data, sitting on a drive. The buyer couldn't find it. The owner had no safe way to sell it. Refinery is built to fix both.

ALT A team spends six months and a few hundred thousand dollars collecting robot demonstrations. Another team already had that data, sitting on a drive. The buyer couldn't find it. The owner had no safe way to sell it. Refinery is built to fix both.

1
2
82
"The best robots of the next decade won't be trained on one company's data. They'll be trained on everyone's, and everyone who contributed should get paid for it." Mohammad Musa, CEO, Deepen AI
"The best robots of the next decade won't be trained on one company's data. They'll be trained on everyone's, and everyone who contributed should get paid for it." Mohammad Musa, CEO, Deepen AI

ALT "The best robots of the next decade won't be trained on one company's data. They'll be trained on everyone's, and everyone who contributed should get paid for it." Mohammad Musa, CEO, Deepen AI

19
Our services rate card is public: $5 per person-hour for common online tasks like data entry, document checks and AI output review. Every team has a lead, QA checks and a weekly report. Start with a two-week paid pilot, no sales call. services.deepen.ai/pricing
Our services rate card is public: $5 per person-hour for common online tasks like data entry, document checks and AI output review. Every team has a lead, QA checks and a weekly report. Start with a two-week paid pilot, no sales call. services.deepen.ai/pricing

ALT Our services rate card is public: $5 per person-hour for common online tasks like data entry, document checks and AI output review. Every team has a lead, QA checks and a weekly report. Start with a two-week paid pilot, no sales call. services.deepen.ai/pricing

1
25
20,000 hours. That's the most generous count of all publicly available real-robot data, from a June study by researchers at NVIDIA, MIT and three other universities. It's roughly what one well-funded lab holds privately.
20,000 hours. That's the most generous count of all publicly available real-robot data, from a June study by researchers at NVIDIA, MIT and three other universities. It's roughly what one well-funded lab holds privately.

ALT 20,000 hours. That's the most generous count of all publicly available real-robot data, from a June study by researchers at NVIDIA, MIT and three other universities. It's roughly what one well-funded lab holds privately.

1
1
22
What pre-verified training data changes: nobody finds out three months in that the calibration was off. Rights gaps don't surface in due diligence. The provenance chain exists before a regulator asks. Engineers work on models, not data cleanup.
What pre-verified training data changes: nobody finds out three months in that the calibration was off. Rights gaps don't surface in due diligence. The provenance chain exists before a regulator asks. Engineers work on models, not data cleanup.

ALT What pre-verified training data changes: nobody finds out three months in that the calibration was off. Rights gaps don't surface in due diligence. The provenance chain exists before a regulator asks. Engineers work on models, not data cleanup.

13
And the newest part: Deepen Refinery, a marketplace for Physical AI training data. Buyers license real-world data with clear terms. Data partners list what they hold and decide who can buy it. Waitlist is open for both: refinery.deepen.ai
20
Then labeling. We annotate multi-sensor data, LiDAR point clouds and camera frames together, so a car or a pallet is the same object in every sensor. Human-verified, because perception models learn exactly what the labels tell them.
22
Start with calibration. A robot or vehicle sees through several sensors: cameras, LiDAR, radar. Calibration tells it exactly where each one sits relative to the others. Get it slightly wrong and every model downstream learns a slightly wrong world.
20
Hi, we are Deepen AI. For nine years we have built the data infrastructure underneath Physical AI: sensor calibration, annotation and validation for robots and autonomous vehicles, across 200+ programs. This account is where we talk about that work in plain terms.
19
Deepen AI retweeted
Mohammad Musa
@MohMusaSep 22
1
2
43
Deepen AI retweeted
Mohammad Musa
@MohMusaSep 22
Physical AI doesn’t just need more data. It needs refined data. Validated. Calibrated. Traceable. Built with provenance. That’s why we built Deepen Refinery: a marketplace for real-world data you can actually trust. refinery.deepen.ai/
2
6
2
12
753
Deepen AI retweeted
Mohammad Musa
@MohMusaSep 22
1/Most of what we built at @deepenai went into ADAS and self-driving. Tens of teams got safer programs on the road with our calibration, labeling, and validation tools. The next generation is robots, humanoids, and world models. They need something automotive never had to solve.
2
2
6
532
Deepen AI retweeted
Mohammad Musa
@MohMusaJul 15
Customers are looking for outcomes and outcomes need credible experts with strong judgement to stand behind it linkedin.com/posts/mohammad_…

When the Model Is a Commodity, Judgment Is the Product | Mohammad Musa

A couple of months ago, I wrote an article about Tech Enabled Services being the new AI moat. You can read it here: https://lnkd.in/eHVQuSpY. In this new article, I explore how the newest AI founda...

linkedin.com
1
2
143
We've closed our Series F round of fundraising at a $15B valuation. Our goal: To bring vehicle intelligence to every moving machine on the planet. Learn more about our vision and road ahead: appliedintuition.com/blog/se… #fundraising #seriesf #startup
10
45
33
196
136,417
AI x Autonomous vehicles: @deepenai Deepen AI is one of the startups that participated at the Beyond Borders competition at the MIT Saudi Business Symposium Mohammad Musa tells us more details:
2
2
299
شركة ذكاء اصطناعي تطمح لدخول السوق السعودي: Deepen AI @deepenai من الشركات المشاركة في مسابقة Beyond Borders يحدثنا عنها محمد موسى في ملتقى الأعمال السعودي في MIT 2025:
4
1
53
5,211
Load more