Copyright yourstory
From factory floors to autonomous vehicles, AI is everywhere. However, there's a critical gap: these systems can only execute what they were trained to do; they can't learn from new experiences in real-time. Once an AI model is deployed to an edge device, such as a machine, camera, or vehicle, it becomes static. When conditions change, different lighting, temperature shifts, or equipment wear, their accuracy drops. The current solution requires shutting down operations, sending data to the cloud for retraining, and reinstalling updated models. This process causes costly downtime. Jagan Teki founded Edgeble AI in Hyderabad in 2022 to solve this issue. "In 24/7 factories, downtime kills productivity. Autonomous vehicles can't wait for the cloud to respond in split-second decisions," he says. Teki has 18 years of semiconductor experience at Qualcomm, Xilinx, and Applied Micro. While deploying an AI defect detection system in 2019, he saw existing solutions couldn't adapt without cloud connectivity. The solution Edgeble AI has developed a self-learning edge AI platform that enables devices to adapt autonomously without cloud connectivity. The company's proprietary algorithm continuously monitors model performance and detects drift and accuracy losses in real-time. It automatically identifies whether issues stem from environmental changes, parameter shifts, or system behaviour changes, then retrains models on the device, corrects machine parameters using small language models, and redeploys updates instantly. The company offers three products: self-learning modules (ready-to-deploy hardware with embedded firmware), self-learning runtime (firmware that upgrades existing edge devices), and a self-learning NPU chip currently in development. The platform supports multimodal AI, including vision, sound, and text processing, and can run large language models at the edge. The system works as an additive layer to existing infrastructure. Rather than replacing neural processing chips, the solution adds autonomy and self-learning capability on top of existing hardware. The platform connects devices in distributed clusters that continuously synchronise and share learning without human intervention, with both online and offline update capabilities. Modules are priced between $200-$850, depending on specifications, the company said. Edgeble AI took 2.5 years from ideation to deployment, with extended R&D to meet industrial demands for AI systems that maintain 24/7 accuracy. Edgeble AI operates in the edge AI hardware market alongside companies like Nvidia (Jetson), Qualcomm, Hailo, and NetraSemi from India. The edge AI hardware market is projected to reach $59.37 billion by 2030, up from $26.17 billion in 2025, at a 17.8% CAGR, according to Mordor Intelligence. Manufacturing and industrial IoT, Edgeble AI's focus area, represents one of the fastest-growing segments as demand grows for systems that adapt and learn autonomously. Commercialisation and expansion The startup focuses on industrial automation and automotive manufacturing, with applications in quality control, predictive maintenance, and process optimisation. Current clients include Amara Raja Batteries for battery production automation, Hero for on-bike intelligence, Patil Group, and Collins Aerospace. The platform also serves the defence, medical devices, and agriculture sectors, where AI systems need to make real-time decisions independently. Edgeble AI is currently bootstrapped and raising seed funding, with plans to deploy 15,000-25,000 edge nodes globally across the industrial automation and automotive sectors over the next two years. The company is also exploring partnerships with existing chip manufacturers, such as Nvidia and Qualcomm, to integrate its self-learning technology into their platforms. Beyond that, Edgeble AI aims to complete its own chip and transition from using third-party processors to its own hardware. The ultimate goal is either licensing the technology to major manufacturers or building a full autonomous edge AI chip independently. The company has filed one patent with three more in progress to protect its self-learning technology. Edgeble AI's strategy is to solve real customer problems first, then build chips based on validated market demand, a reverse approach to traditional semiconductor development. "We want to prove our technology solves real problems in the field before we invest heavily in building our own chip. That ensures we're addressing actual industrial needs, not just theoretical applications," Teki says. Edgeble AI is part of YourStory’s Tech30 cohort—a selection of India’s most promising startups of 2025—unveiled at TechSparks Bengaluru. (Edited by Affirunisa Kankudti)
