Big ideas,
built in a nutshell.

We are taking the games we grew up playing, in courtyards, living rooms, school grounds and around family tables, and rebuilding them as modern hybrid casual experiences for mobile. The goal is not to simplify away what made those games special, but to preserve their character, competition and familiarity while making them more approachable, expressive and enjoyable on a phone.

That means combining the rules and cultural memory of traditional games with today's mobile craft: tactile interactions, satisfying animation, real physics, distinctive characters, fast onboarding and the kind of visual feedback players now expect from polished mobile games. We want each game to feel immediately familiar to someone who grew up with the original, while still being inviting to someone discovering it for the first time.

Two titles are currently in development.

The first reimagines chess around one problem we have seen repeatedly: for a new player, chess can feel abstract before it ever becomes exciting. Thirty-two pieces appear on a board, many represented by similar symbolic forms, and beginners are expected to learn what every piece is, how it moves and why it matters before they can really enjoy the game.

We are changing the presentation without changing the game.

Every chess piece becomes a recognizable character with its own identity, movement and animation, designed so that its role becomes easier to understand at a glance. Knights still move like knights. Rooks still command files and ranks. Bishops still cut across diagonals. Every legal move, rule and strategic possibility remains intact.

Underneath, it is still real chess: tournament-honest and mechanically unchanged. What changes is the way the board communicates with the player.

Movement becomes more expressive. Captures have physicality. Pieces feel alive. The relationship between character and function becomes easier to understand, particularly for children and first-time players. Instead of asking new players to decode the board before enjoying it, the game helps the board explain itself.

The ambition is simple: preserve everything that makes chess deep enough to last centuries, while removing some of the visual and emotional friction that stops people from getting through their first few games.

The first title is planned to ship in the coming months.

The second takes the same philosophy into another traditional game we grew up with. That one is ours to talk about later.

First title in a few months

Nutshell-Robotics brings together everything we build that moves: swarm microrobots for disaster sites and greenhouses, and companion robotic toys for play and learning. All of it is designed in our studio, from the circuit board up.

Swarm robotics

We are developing bio-inspired swarm microrobots that crawl through rubble, climb broken structures, move through flooded spaces and, in time, fly across terrain before deploying into spaces too small, unstable or dangerous for people.

Instead of one large, expensive rescue robot, we are working toward many small, deliberately low-cost machines that spread through a site together. Walkers squeeze through rubble and voids. Climbers inspect walls, pipes and vertical shafts. Swimmers enter flooded rooms and culverts. Fliers cross open ground and carry sensing where roads have disappeared.

Each robot is a small sensing node: thermal imaging that finds heat where a camera sees nothing, on-board vision that flags a face, a hand or a light, microphones that listen for tapping, and the micro-motion radar we are exploring to catch breathing through rubble. Recognition and navigation run on the robot itself, because underground there is no link home, and what one robot finds, the swarm shares. It is closer to an ant colony than a fleet of remote-controlled machines.

With a different payload, the same swarm works our greenhouses. Units carrying CO2, temperature, humidity and light sensors move through the crop and map conditions that change from one row to the next, and some carry a small pair of jaws to take off a mouldy leaf or a tomato side shoot, among the most labour-intensive jobs in a glasshouse. Where to cut, and how to keep those jaws clean between plants, is part of what we are researching.

Companion robotic toys

Our companion robotic toys are compact enough for a desk and built to be played with, talked to and learned from. Each one walks on two legs and keeps its own balance, a skill it learns in simulation, falling over thousands of times before it takes a single real step. An expressive head turns, tilts and nods to show what it is paying attention to. A camera recognises faces, microphones pick out who is speaking, and a speaker lets it respond. As much of the AI as possible runs on the robot itself, so a child's face and voice never need to leave the room.

Unlike most small robots, ours have arms: two small human-like arms that wave, point, shrug and pick up light objects, so the robot communicates with its whole body, the way children do. It is a companion first and a teacher second.

It dances, plays games and answers your voice, and it handles simple tasks too, such as fetching and carrying light objects or setting timers and reminders. It reacts to what happens around it: it looks up when someone walks in, follows a moving hand, flinches at a sudden noise and shows when it is happy, curious or sleepy. Every move it makes is code you can open and change, starting with simple blocks and growing into real programming.

We design the printed circuit boards and 3D print the parts in-house, so electronics and mechanics come from the same bench and either can be changed whenever a design needs to improve.

Further ahead, the same workshop will develop advanced fruit pickers, drawing on what the swarm robots are teaching us about sensing and cutting in the crop.

Nutshell-Air Mobility is our own eVTOL: a lightweight, single-seat electric aircraft that takes off and lands vertically and flies low, close to the landscape rather than high above it. It is designed to be flown by the person on board rather than a professional pilot: intuitive controls and automated flight assistance mean ordinary people can learn to fly it after a short training course.

Ease of flying comes from the control system. A fly-by-wire flight computer fuses inertial, GPS and barometric data hundreds of times a second and sets the thrust of every motor through our own motor controllers. It holds a stable hover hands-off, levels the aircraft in gusts and runs automatic take-off and landing. One stick sets direction and speed, a geofence keeps it inside a safe area and below a safe height, and if the battery runs low or a fault is detected, it lands itself.

Safety is designed in from the frame up. The aircraft carries more motors than it needs to stay in the air, so losing one does not mean losing the aircraft, and a whole-aircraft parachute is part of the design for the moment everything else has failed.

We build the hardware ourselves. The airframe is designed and 3D printed in our studio, so a new arm or fairing goes from the screen to the test stand in days rather than months, with lattice interiors that keep the strength and lose the weight. We design our own PCBs and electronics, including the motor controllers, and assemble the flight computers and battery packs in-house. When a part needs to change, nothing waits on a supplier.

An electric aircraft that can land almost anywhere opens up a different kind of travel.

Recreation and entertainment: short scenic flights over valleys, lakes and ridgelines, as an experience in themselves.

Sightseeing: mountain views that today take a week of trekking or a long road journey.

National parks: designed to be much quieter than a helicopter, with no exhaust over the wildlife and landscapes that need protecting.

Air ambulance: reaching injured trekkers and remote villages in minutes, landing in clearings too small for a helicopter, and flying in medicines and first-aid supplies where no road or vehicle can go.

It is an early-stage design. Between a concept and a person in the sky stand years of prototypes, flight testing and approval from aviation authorities, and the rules for light aircraft differ from one country to the next. We intend to do every step properly, in every place we fly.

Hearing loss is easy to overlook because nobody can see it. We have seen it in people close to us: someone who used to lead the conversation at dinner starts smiling and nodding, misses the punchline, stops answering the phone and, little by little, stops going out. The research says the same. A systematic review led by Johns Hopkins researchers found that, in most studies, hearing loss in older adults was associated with a higher risk of loneliness and social isolation. The 2024 Lancet Commission on dementia estimates that hearing loss accounts for about 7 percent of dementia cases worldwide, making it one of the largest modifiable risk factors.

Treating it appears to help. A University of Michigan study of more than 100,000 older adults with hearing loss found that those who got a hearing aid had an 18 percent lower risk of a dementia diagnosis, an 11 percent lower risk of depression or anxiety and a 13 percent lower risk of fall-related injuries over three years. A University of Southern California study found that adults with hearing loss who wore hearing aids regularly had a 24 percent lower risk of death than those who never did. And in the ACHIEVE trial led by Johns Hopkins, hearing care did not change cognitive decline overall but slowed it by almost half in older adults already at higher risk. These are associations and early results, not guarantees, but they all point the same way: hearing well keeps people connected.

The scale is enormous. The World Health Organization estimates that around 1.5 billion people live with some degree of hearing loss, about 430 million of them at a disabling level, a number expected to approach 700 million by 2050. Nearly 80 percent live in low- and middle-income countries. Most could be helped: commonly cited estimates put the share of hearing loss that hearing aids can treat at around 90 percent. Yet only about 17 percent of people who could benefit from a hearing aid use one, and among adults in low- and middle-income countries the figure falls below 3 percent.

Price is the barrier. In the United States an average pair costs around 4,700 dollars, and in a country like Nepal a good pair can cost more than a family earns in several months. Something that should be ordinary has become a luxury. Our goal is simple to state and hard to achieve: hearing aids with the sound of premium devices at a price ordinary families can pay, so that far more of the people who need one actually wear one.

Speech clarity. A hearing aid's job is not to make everything louder but to make speech understandable. Our processing splits sound into many frequency channels, fits the gain to each person's own hearing profile, and runs a deep neural network trained on a large database of real-world recordings, from busy streets and restaurants to wind, traffic and crowded rooms, so it learns to separate a voice from its surroundings rather than relying on hand-written rules.

Noise and feedback cancellation. Directional microphones and beamforming focus on the person you are facing, while noise reduction lowers steady background sound. A phase-cancellation algorithm tracks the whistle that appears when amplified sound leaks back into the microphone and cancels it with an inverted copy of itself before you hear it, so the aid can deliver more gain without whistling.

Sensors that understand the moment. Sound alone cannot tell a hearing aid what you want to hear. Motion sensors follow head and body movement, while acoustic sensors measure sound level, signal-to-noise ratio and whether a conversation is going on. Combining those four inputs, the aid can tell whether you are turning toward someone, walking down a busy street or sitting in a quiet room, and it narrows its focus on a conversation or opens up to your surroundings to match.

Small, all-day and connected. All of this has to fit behind or inside the ear, so miniaturisation shapes every decision, from the chip and antenna layout on our own PCBs to where the microphones and receiver sit. A rechargeable lithium-ion cell is sized to last a full day including streaming, and Bluetooth LE Audio sends calls, music and television straight to the ear, including the new broadcast audio now arriving in public venues.

It is early work. A hearing aid is a medical device, and before ours reaches anyone's ear it has to pass clinical testing and regulatory approval in every market where it is sold. We intend to do that properly, because the people we are building it for deserve a device that is both affordable and excellent, not one or the other.

iHatePDF.com started with a simple frustration: working with PDFs should not require expensive subscriptions, complicated software, account creation, or a different website for every small task. So we built the document platform we wanted to use ourselves: fast, straightforward, privacy-conscious, and available to anyone from a browser.

Today it brings 35+ document tools into one place, covering the everyday work people normally need several applications to handle. Merge and split PDFs, compress files, convert between PDF, Word, PowerPoint, Excel and images, run OCR on scanned documents, sign and annotate files, redact sensitive information, organize pages, compare documents, translate PDFs, and more.

For an even simpler workflow, documents can also be handled through chat with Buddy, our built-in assistant: upload a file, ask for what you need (convert this to Word, compress this PDF, help me understand this document) and the platform connects the request to the right tool.

At the centre of the product is Editly, our own document editor built for working directly with PDF and DOCX files. Instead of treating a PDF as a collection of static pages, Editly lets you work with the document itself: rewrite existing text in place while preserving its original typography, add or reposition images and shapes, fill forms, draw, highlight, annotate, and place signatures directly onto the page.

For users who want additional privacy, an optional browser mode performs supported processing locally on the device instead of sending the document away for processing.

No account required for everyday use. No watermarks added to your finished files. No artificial export limits. Just a growing set of professional document tools built to make one of computing's most frustrating formats considerably easier to work with.

We are building sensor-driven hydroponic and aeroponic greenhouses, for Nepal first and eventually for anywhere that needs more fresh food from less land and water.

The problem is bigger than putting vegetables under glass. Growing reliably means managing a whole living system: the air around the plant, the water at its roots, the nutrients it takes up, the light it receives, the disease developing before anyone can see it, and whether the crop will still have a market when it is ready to harvest.

So we bring all of it into one closed loop. Plants grow without soil in recirculating hydroponic and aeroponic systems, and whatever the plant does not use stays in the loop instead of draining into the ground. Our own sensor network continuously measures what actually decides how a crop develops: light, air and water temperature, humidity, CO2, pH, electrical conductivity, nutrient strength and what the roots are drinking.

Those readings are not there to make a dashboard look intelligent. They drive the greenhouse. Pumps regulate feeding. Misters control the root environment. Fans manage heat, humidity and airflow. Lamps supplement what the sun cannot give. Dosing systems correct pH and concentration. The grower sets the conditions the crop needs and the control system holds the house close to them, day and night, without anyone walking every row to turn a valve.

Over time we want it to do more than hold conditions. We want it to understand what the plants are doing inside them. Cameras and environmental data together look for disease and stress before they spread. Growth models follow a plant through its development instead of treating every day the same. Nutrition models learn from what the crop actually takes up, so feeding follows the plant rather than a calendar. The aim is a house that can answer what normally needs an experienced grower standing beside the crop: is this plant developing normally, is something starting to go wrong, does it need more water, less nitrogen, different light.

One more signal matters: the market outside. A perfect tomato is only useful if somebody needs it the week it is picked, so our market models combine crop cycles with prices, seasonality and demand to inform what to plant, when, and when a harvest is likely to be worth most. That counts especially in Nepal, where mountain geography and seasonal production make fresh supply uneven, and vegetables that are cheap one month are scarce the next.

Closed-loop growing also changes the economics of the land itself: substantially more food from a given footprint on a fraction of the irrigation water, and more still from stacked and vertical systems. We do not want to solve water and land by ignoring energy, so the design leans on natural light, efficient pumps, renewables and climate strategies suited to where the greenhouse is actually built. The point is not to copy an energy-hungry European glasshouse into Nepal, but to take the science of precision horticulture and rebuild it around local economics, materials and conditions.

It shares a stack with our robotics work. The same low-power sensor boards, cameras, thermal sensing, edge intelligence and IoT that help a robot read an unknown space help a greenhouse read a crop. Fixed sensors report single points; mobile sensing can eventually move between plants, inspect leaves and map temperature and humidity, giving the house eyes that travel through the crop instead of staying bolted to a wall.

The work is at R&D stage: building and testing the sensing platform, control loops, nutrient systems, dashboards and plant models against real growing conditions. Next come pilot greenhouses in Nepal. The ambition is not simply to automate farming. It is a greenhouse that knows what its plants need, uses only what they need, spots trouble early and helps growers produce the right food at the right time, on less land, with less water and less guesswork.