doubleAI
doubleAI is a frontier AI lab focused on the hardest problems in science and engineering. It searches the space of ideas and applies formal verification so that expertise can scale without limits.
At a Glance
- AI labs and companies with large GPU/CPU training or inference bills
- Software and infrastructure engineering teams
- Data scientists and researchers
- Scientific and mathematical research organizations
- +2 more
AI Tools by doubleAI
(1)
Maxwell
Runtime Code Optimizer with Proofs
Discussions
No discussions yet
Be the first to start a discussion about doubleAI
Latest News
doubleAI launches Maxwell, the first Code Execution Model, to optimize software while it runs
Wall Street Journal reports NVIDIA-backed doubleAI's model cut an early customer's compute costs by 67%
Double Agent reported a new extremal graph-theory result, with a formally verified proof
doubleAI named among LinkedIn's Top 15 Israeli Startups for 2026
Products & Services
A Code Execution Model that lives inside running programs, observes runtime data and hardware, searches for faster algorithms and machine code, and replaces code with formally verified semantically equivalent binaries. It targets CPU and GPU workloads without changing source code.
doubleAI's expert-level AI research agent for deep scientific and engineering problems, including formal mathematics, hardware design, quantum-circuit design and algorithmic discovery.
The company's broader research approach: domain-specific systems that search deeply, learn from experience and use formal verification to produce reliable expert-level results in hard scientific domains.
Market Position
doubleAI positions itself between coding agents and compilers: instead of merely writing source code or applying prewritten compiler transformations, its Code Execution Models observe running programs, search for algorithmic and low-level improvements tailored to each machine and data stream, and formally prove equivalence. Relevant comparison points include torch.compile/PyPy and traditional JIT compilers, static superoptimizers such as STOKE, and coding-agent platforms; doubleAI differentiates through runtime whole-function/algorithm rewrites plus machine-checked correctness.
Leadership
Founders
Prof. Amnon Shashua
Computer-vision professor and founder/CEO of Mobileye; also founded or helped launch CogniTens and has been involved with AI21 Labs and OrCam. CEO of doubleAI.
Prof. Shai Shalev-Shwartz
Machine-learning professor and Mobileye CTO; co-founder and CTO of doubleAI.
Dr. Gal Beniamini
Researcher and member of the founding team, with a background spanning computer science, mathematics, physics and biology research.
Prof. Yoav Levine
Research scientist and former AI21 Labs co-Chief Scientist; co-founder of doubleAI.
Prof. Or Sharir
Research scientist and former AI21 Labs research scientist; co-founder of doubleAI.
Dr. Noam Wies
Algorithm specialist and former OrCam researcher; co-founder of doubleAI.
Executive Team
Prof. Amnon Shashua
Co-founder & CEO
Computer-vision professor and Mobileye founder/CEO; entrepreneur behind Mobileye and former founder/investor associated with CogniTens, AI21 Labs and OrCam.
Prof. Shai Shalev-Shwartz
Co-founder & CTO
Machine-learning professor and Mobileye CTO; researcher in learning and reasoning.
Founding Story
AA-I Technologies/doubleAI was founded in August 2023 in Jerusalem around the idea of integrating domain-specific expertise with precise, critical reasoning comparable to top human experts. The founding team combined Shashua and Shalev-Shwartz's Mobileye experience with researchers from AI21 Labs, OrCam and academic research, initially pursuing Artificial Expert Intelligence for difficult scientific domains.
Business Model
Revenue Model
Outcome-based pricing: doubleAI plans to charge customers a share of the compute savings Maxwell delivers, rather than simply charging a conventional fixed software subscription.
Target Markets
- AI labs and companies with large GPU/CPU training or inference bills
- Software and infrastructure engineering teams
- Data scientists and researchers
- Scientific and mathematical research organizations
- Hardware and semiconductor design teams
- Quantum-computing and cryptography researchers
- Reducing inference and data-processing compute costs for AI labs and other compute-intensive organizations
- Optimizing Python, pandas, PyTorch, graph algorithms and GPU kernels
- Accelerating video-generation pipelines such as LTX-2.5
- Speech-to-text and other CPU/GPU inference workloads
- Scientific and mathematical discovery where results can be formally verified
- Hardware design and optimization
- Lightricks / LTX: Maxwell accelerated LTX-2.5 4K video generation 1.87x, from 250 seconds to 134 seconds per clip on two B200 GPUs