AI continues to dominate the startup landscape as several new combinator-backed ventures emerged on Friday. Among them is Stellon Labs, which develops “ultra tranny” frontier models that can be run on almost any device, except for the need for a GPU.
It was founded by Rohan Joshi and Divom Gupta, researchers at Carnegie Mellon University AI. Due to these requirements, most models are not accessible on edge devices such as smartphones, laptops, robots, and embedded systems.
Stellon Labs aims to change it by creating compact, high-performance models of voice, language and video intelligence. In particular, it is built to run efficiently on everyday hardware without sacrificing quality.
Read: NOX Metals aims to redefine US manufacturing in its automated metals factory (
Another notable Y-combinator-assisted AI startup uses frizzy, to evaluate handwritten math assignments and worksheets. Founded by Abhay Gupta and Shyam Sai, the platform allows teachers to score thousands of worksheets and save hundreds of hours manually by trying to save money.
Frizzle aims to use AI to solve the problem of teachers being overworked, freeing up time spent grading assignments, allowing them to focus on support students. Startups claim to provide accurate grading, actionable insights and student-friendly feedback.
Gupta previously worked as a product manager at Coinbase, where he helped generate $50 million in incremental revenue, but Joshi is a machine learning engineer at Microsoft, patented LLM applications and co-founded Midwest Math Circle.
Read: Vulcan Technologies brings AI to the heart of American law (
Finally, there is Ondeck, an AI startup that focuses on video analytics. Founded by Alexander Dungate and Sepand Dyanatkar, the company aims to address the challenges of building computer vision models, which typically involve months of engineering efforts to collect, train and deploy data. Traditional models struggle to generalize to a wide range of camera setups, workflows and environments, making it almost impossible to obtain sufficient training data for a given task.
Ondeck is working on this issue with the Vision Language Model (VLMS). Its vision engine can generalize across tasks without the need to train data, and has already analysed thousands of hours of footage, including autonomous surface vehicles, robotic research, security systems, offshore oil and gas monitoring.
