For millions of people with visual impairments navigating bustling streets or reading printed documents becomes a quiet challenge. While several tools like Microsoft Seeing AI and Google Lookout offer promising solutions, most existing assistive technologies are limited in real-time scene understanding, emergency response, and require constant internet connection.
Akumen Artificial Intelligence (Akumen AI) develops computer vision and AI technologies for assistive devices. The technology is integrated into wearables such as INSIGHT, Akumen AI’s visual aid device designed as a pair of sunglasses. The device processes visual information in the user’s surroundings and translates it into real-time spoken audio INSIGHT’s core assistive capabilities also include text reading, object detection, and banknote-denomination recognition.
Conversations with the CEO
We spoke with Akuvan Thiyam, CEO of Akumen AI, to discuss the technology and vision driving INSIGHT. In our conversation below, he reflects on the engineering behind the wearable device and how the company balances responsible IP protection over their innovations.
Q: What originally inspired you to start Akumen AI, and what problem did you want the company to solve?
A: Akumen AI was inspired by my late uncle, Dr. Ferdinand Mugusi, who was diagnosed with retinitis pigmentosa, a progressive condition that gradually took away his vision. Despite being a highly accomplished medical professor, he had to depend on assistive solutions that were often bulky, internet-dependent and unable to provide meaningful guidance in everyday situations. Before he passed away, we were fortunate to test several early iterations of our INSIGHT smart glasses with him. His experiences and feedback helped shape both the product and the purpose of Akumen AI.
Our goal is to help blind and low-vision people better understand their surroundings through timely, contextual cues delivered through natural speech, audio or haptic feedback, without overwhelming them or adding unnecessary cognitive load. We want technology to function as a quiet, intuitive companion that supports independence, confidence and dignity in everyday life.
Q: What makes INSIGHT different from other AI-powered assistive devices currently available to visually impaired users?
A: What distinguishes INSIGHT is its practical, user-centred architecture. Our user research indicated that many prospective users already use compatible smartphones, creating an opportunity to use the phone’s computing capabilities instead of duplicating all the processing hardware within the glasses. INSIGHT is therefore designed as a lightweight wearable that captures user-initiated information through its camera and sensors, communicates with a connected smartphone for processing, and delivers guidance through speech, audio or haptic feedback.
Core functions such as text reading, object detection, banknote-denomination recognition and selected environmental-awareness features are designed to run locally on a compatible smartphone wherever possible. Certain optional advanced features may use cloud services, while the essential offline functions remain available without continuous internet access. Local processing can improve responsiveness, reduce dependence on connectivity and limit the amount of user data transmitted off the device.
Separating the wearable interface from the main computing platform is intended to reduce the glasses’ weight, heat generation, power consumption and overall cost. It also allows us to improve supported functions through software updates without regularly replacing the wearable hardware. INSIGHT has been designed from the outset with blind and low-vision users in mind, with particular attention to contextual assistance, Indian-language support and feedback intended to provide useful information without creating unnecessary cognitive load.
Q: What were the biggest engineering challenges in combining computer vision, AI, cameras, sensors and audio into a lightweight wearable device, and what decisions have been most important in making advanced assistive technology more accessible?
A: The greatest engineering challenge was integrating cameras, sensors, audio, wireless communication and power management into a device designed to remain lightweight and comfortable for everyday use. This required several hardware iterations involving PCB layout, thermal management, battery placement, antenna performance, weight distribution and mechanical enclosure design. Each iteration involved redesigning and fabricating the boards, assembling the device, testing its performance and incorporating the resulting improvements.
Firmware integration was equally demanding. Many hardware components did not have software libraries that worked together reliably within our architecture. Our team therefore developed and optimised device-specific drivers and integration layers while using appropriately licensed components where suitable.
The smartphone application serves as the computational and coordination layer of the system. It must reliably receive information from the glasses, identify the requested function, run the appropriate processing pipeline and return useful feedback with minimal delay. We develop and optimise task-specific AI pipelines for this purpose, fine-tuning and quantizing wherever applicable. The models must meet defined performance targets while remaining compact and efficient enough to run locally, requiring careful trade-offs among accuracy, model size, inference speed and power consumption.
An equally important challenge was the Human Computer Interaction. We had to determine when to use speech, short audio cues or haptic alerts; how much information to provide; and how to avoid unnecessary cognitive load. These decisions have been refined through repeated testing and feedback from blind and low-vision users.
The most important accessibility decisions were to use a companion smartphone for intensive processing, prioritise offline operation for essential functions, keep the hardware architecture modular and involve intended users throughout development. Together, these choices are intended to reduce device weight, connectivity dependence and cost while making the experience more practical for everyday use.
Q: Which technological innovations are the most novel or important to protect through intellectual property?
A: The most important innovation to protect is the specific system architecture that enables a wearable device to collect multimodal sensor information, coordinate processing across edge devices and servers, and return timely contextual feedback through audio or haptic cues. This includes the way visual, audio, depth and motion data can be processed to support functions such as object and obstacle recognition, contextual assistance, navigation-related feedback, fall detection and emergency notifications.
This architecture is covered by U.S. Patent No. 12,693,732, titled “System and method for providing real-time insights to user,” which was granted to Akumen Artificial Intelligence Private Limited on July 28, 2026. The patent protects a particular claimed combination of wearable sensing, power and communication management, AI-based processing, and multimodal feedback. It does not claim ownership over AI, computer vision or assistive technology generally; its value lies in the specific architecture and workflow described in the granted claims.
We would not want intellectual property to restrict the wider development of assistive technology. Accessibility is too important for any one company to control. Open AI models, interoperable tools, developer platforms and partnerships with the blind and low-vision community can enable researchers and organisations to build new experiences beyond what a single company could create.
We therefore want to balance responsible IP protection with open standards, interoperable tools, research partnerships and appropriate access for developers. We are open to licensing and collaboration with organisations that share our commitment to accessibility. Our objective is to protect the integrity and sustainability of INSIGHT while enabling a broader ecosystem of affordable assistive solutions. Ultimately, our success should be measured not by how much technology we control, but by how many visually impaired people gain greater independence through it.
Q: How do you see AI, robotics and wearable technology coming together to support people with different disabilities?
A: At Akumen AI, we see the future of assistive technology as a user-controlled network of intelligent devices rather than a collection of isolated products. With the user’s permission, smart glasses, smartphones, environmental sensors, smart canes, wheelchairs and robotic systems could exchange only the contextual information necessary to provide coordinated assistance.
Methods related to those used in autonomous systems: including computer vision, sensor fusion, localisation, mapping and obstacle detection could inform the development of smaller and more power-efficient assistive devices. However, assistive products would require independent testing and validation for pedestrian environments and the specific needs of their intended users.
AI, robotics and wearables could also support people with other disabilities by presenting information or controls in the form most useful to each individual – for example, captions for someone with hearing loss, alternative communication interfaces for someone with a speech impairment, or accessible controls for a mobility device. These applications cannot be approached as a single solution for every disability; each must be co-designed and validated with the relevant disability community. Our vision is a standards-based, privacy-conscious and user-controlled ecosystem that supports greater choice, participation and independence.
Akumen AI’s key patent
Providing real-time insights to user
Visually impaired individuals face daily navigational hazards which increase their risk of falls and unassisted emergencies both indoors and outdoors. Current assistive technologies often experience technical limitations, such as delayed processing speeds, limited environmental recognition, and challenges in integrating multiple sensors to aid in navigation.
U.S. Patent No. 12,693,732 gathers comprehensive environmental data using a network of wearable devices. This network includes a pair of smart glasses, a headgear, a headband, a chest-mounted device, a wearable pendant, a smart watch, a wrist worn device, a cane, shoes, and earbuds. Integrated sensors capture visual data via cameras for scene recognition and object detection along with microphones which process audio data for speech recognition or sound classification. On the other hand, point-cloud data is collected via light detection and ranging (LiDar) while motion and orientation data is captured via an Inertial Measurement Unit (IMU). Additionally, the device detects obstacles through an obstacle detection unit by computing the distance between the user and an obstacle. It then produces dynamic warning sounds as obstacles get closer.
To predict falls, the system utilizes gait analysis to detect abnormal walking patterns of the user. Upon detecting abrupt motion changes, the system notifies one or more preregistered emergency contacts with the user’s exact location. Additionally, the device delivers real-time haptic feedback to the user through distinct vibration signals to raise situational awareness. The different vibration patterns help users distinguish between obstacles and directional prompts to navigate safely on their own.
A dual-tier architecture powers the system to minimize processing delays. Time-sensitive tasks, such as immediate obstacle recognition, are executed locally on the user’s cellphone using light artificial intelligence (AI) models. Meanwhile, more complex tasks like answering conversational questions are routed to cloud-based Large Language Models (LLMs). To ensure these models run efficiently on the device without draining the battery, the system employs federated learning and data compression to ensure that core safety features remain fully operational even without an active network connection.
This patent entitled “System and method for providing real-time insights to user” was filed on July 25, 2025 and granted on July 28, 2026. Thiyam Akuvan, Gyanendro Khomdram, Thoudam Kheljeet Singh, Oinam Roshan, and Phijam Nongthanganba are listed as the inventors.
Akumen AI’s vision for assistive technology
We wrapped our conversation with Akuvan with a discussion on the next five to ten years of AI-powered assistive technology and Akumen AI’s intended role.
According to Akuvan, the industry is expected to shift “from isolated features into more continuous, personalised and context-aware support systems.” He noted that lightweight wearables combining various sensors and spatial intelligence could assist users in understanding their surroundings, provided there is “appropriate consent and privacy controls.”
Akuvan stated that Brain Computer Interfaces (BCIs) “may eventually provide an additional interaction channel for people who cannot reliably use speech, touch or conventional physical controls.” Rather than reading thoughts, he explained that these systems would “decode limited, task-specific neural signals.” He emphasized, however, that BCIs are an emerging field meant to complement rather than replace established tools. Ultimately, Akuvan firmly asserted that “clinical validation, safety, cybersecurity, mental privacy, informed consent and user autonomy must remain central to their development.”
Looking ahead, Akuvan noted that the company’s broader objective is to prevent advanced assistive technology from being limited to exclusive markets, ensuring it instead becomes “affordable, interoperable, privacy-conscious and genuinely useful in everyday life.”
