Latent AI develops software that compresses and optimizes AI models—claiming roughly 10x compression—so they can run efficiently on devices without constant cloud connectivity. Its technology is targeted at use cases like local face-detection in cameras, instantaneous voice interfaces, and augmented-reality goggles. The company sells access to its tools on a subscription basis and customers deploy the tools on-premises. Latent AI reports a small but growing customer base, including an unnamed steel manufacturer using its tech in AR goggles and, reportedly, a U.S. government customer. The startup positions itself against TensorFlow, chip vendors like Xilinx, and rivals such as OctoML and Deeplite, arguing competitors focus on either compression or compiling rather than both. The team is roughly 15 people and is working to scale product development and customer deployments amid competitive pressure in edge AI optimization. Latent AI is a startup spun out of SRI International that builds the Latent AI Efficient Inference Platform (LEIP) to make edge AI workloads more efficient. LEIP is offered as standalone modules (compressor and compiler) or as a fully integrated system, and the company is launching LEIP Adapt to dynamically manage workloads. Using its proprietary compression and compilation process, Latent AI says it can compress library files by 10x and achieve 5x lower latency while using less power. Practical use cases include battery-powered smart doorbells that run in low-power mode until a larger model is needed, and the ability to offload to the cloud when required. The company was founded by CEO Jags Kandasamy and CTO Sek Chai, drawing on Kandasamy’s prior edge work at OtoSense and Chai’s SRI research on power‑aware compute. Latent AI has attracted interest from automotive firms, consumer product companies (including a camera and a hearing aid maker), a major telco, and a large CDN provider, and has raised a $6.5 million seed round. Latent AI develops a modular, fully integrated workflow platform for edge AI that trains, quantizes and deploys neural networks. Its platform is designed for the adaptive edge and includes LEIP Compress, a quantization optimizer supporting post-training and training-aware quantization, and LEIP Compile, which optimizes neural network processing for hardware targets. The company is led by co-founders Jags Kandasamy (Co-Founder & CEO) and Sek Chai (Co-Founder & CTO). Latent AI is a portfolio company of SRI Ventures, a division of SRI International. The article does not disclose operating metrics or revenue.

Total raised

$29M

Funding rounds

3

Latest round

Series A

Latest activity

Aug 2021