GreenVisions PH: a soil test that answers in minutes instead of months
A Philippine social enterprise built a field device that reads soil nutrients in seven minutes and prints a crop-specific fertilizer plan on the spot.
Read →
Notes on building fast, scalable web applications with Next.js and headless CMS — the decisions and trade-offs behind real client work.
A Philippine social enterprise built a field device that reads soil nutrients in seven minutes and prints a crop-specific fertilizer plan on the spot.
Read →A four-founder New York startup built a custom chip that runs real-time AI source separation inside a hearing aid, tackling the cocktail-party problem old presets never solved.
Read →AI's real infrastructure bottleneck is the years-long wait to connect to the grid. Atomarine skips the queue entirely by building floating, modular data centers cooled by seawater, with nuclear power on the roadmap.
Read →An Australian startup trains drones with reinforcement learning to muster cattle the way an experienced stockman would, backing off the moment an animal shows stress.
Read →A small Maryland startup uses synthetic diamonds and machine learning to map buried currents inside 3D chips, replacing the slicing and grinding that chip inspection has always required.
Read →A small startup took the safety engineering behind wearable rehab robots and pointed it at a beauty task done one hair at a time, inside real Ulta and Nordstrom stores.
Read →A small Stanford-founded startup builds earbuds whose AI model isolates the wearer's own voice instead of blocking noise, catching speech below a whisper or across a loud room.
Read →A small Hong Kong robotics startup turns touch into an image its robots already know how to read, packing over 110,000 sensing points into a single fingertip.
Read →A small Cleveland startup founded by former NASA acoustics engineers uses inaudible low-frequency sound waves to detect and suppress fires in seconds, without water or chemicals.
Read →A small Wisconsin beekeeping company built a solar-powered hive lid that listens for a colony about to swarm, days before a beekeeper would notice.
Read →How AI's technical evolution and the regulation trying to keep up with it actually fit together — from the 2017 Transformer to the EU AI Act's 2026 enforcement.
Read →A Turkiye-based startup runs NLP over messy hospital notes to surface rare-disease clues that no single doctor could ever read at scale.
Read →Cosa è già in vigore, cosa è stato rinviato e cosa fare subito, aggiornato alle scadenze reali di fine agosto 2026, per chi sviluppa e per chi usa l'IA.
Read →What's actually in force, what got postponed, and what to do now, updated to the real deadlines as of late August 2026, for developers and businesses alike.
Read →An Estonian-Polish startup reads heart rate, blood pressure, and breathing rate from a 30-second phone camera scan, and its Medical SDK just earned CE certification as a Class IIa device.
Read →A Poland-based startup built an agentic OS that listens to a technician on site and turns the visit into a priced, compliant, billable job before the van leaves.
Read →A Geneva spinout packs a stethoscope, pulse oximeter and thermometer into one pocket device, then lets an AI model read lung sounds instead of a trained ear.
Read →A University of Pennsylvania spinout runs modular robotic labs that grow twenty kinds of human tissue, generating the causal biological data that today's AI drug-discovery models are missing.
Read →A Tel Aviv startup mounts thermal and day cameras above the bridge so ships can spot small, unlit vessels that radar and AIS miss, and Lloyd's Register just ran a real Mediterranean trial to check the claim.
Read →A five-person YC startup built WaveSight, a radio-wave camera that turns hidden pipes, wiring and studs into a millimeter-accurate 3D map, without touching the wall.
Read →A Kochi startup's underwater robots inspect dams, ship hulls and pipelines for industry, then survey coral reefs in the Andaman and Lakshadweep islands for marine scientists, using the same AI vision layer.
Read →A small German-Dutch neuroscience startup builds a brain-wave headset that flags the instant a person registers error or surprise, adding that signal to the training data robots learn from.
Read →A London and Cambridge startup is building processors where computation runs both forward and backward, so the energy that would normally leak out as heat gets reused for the next calculation.
Read →A three-person NC State spinout built a leaf sensor that reads the gas a plant emits under stress, catching disease about a week before any visible symptom.
Read →A small California startup built a prosthetic arm whose AI neural interface reads muscle signals through the skin, giving amputees individual finger control at a fraction of the usual cost.
Read →A chip design startup built an agentic harness that turned one written spec into a working RISC-V CPU core in twelve hours, no human writing the register-transfer code.
Read →A small startup is teaching power grids to recognize the electrical signature of a wildfire before it ignites, not just cut power after it does.
Read →A Purdue spinout replaced expensive survey vehicles with a low-cost sensor kit and computer vision that scores every crack and pothole by depth, and has already mapped over 3,400 miles of Indiana roads.
Read →A small Dutch startup built a free-fall sorting machine that pulls loose batteries out of household waste at 96 percent purity, using a camera that tracks each falling object instead of a conveyor belt.
Read →A Swiss team of ex-NASA and ETH Zurich engineers built modular underwater robots that dock, recharge, and stay at sea for six months, turning annual ship inspections into a continuous digital twin.
Read →A self-funded Nigeria-and-UK team builds a bidirectional ASL translator that runs on hand landmarks instead of raw video, so it keeps working offline and under 100ms.
Read →A Nigerian startup builds a handheld scanner that spots fake pills in seconds. When the internet failed during a live demo, the fix reshaped the whole product.
Read →A tiny Irish-founded startup skips the teach pendant entirely. Forge scans each part to sub-millimeter accuracy first, then lets the welding arm plan its own path, closing the gap that leaves 10% of welded parts needing rework.
Read →A six-person Doha team built a phone app that identifies a hidden infestation by sound alone, catching a beetle larva weeks before any leaf shows damage.
Read →Three Hamburg graduates built LV01, a modular wearable that turns a camera view into a spoken scene description. The trigger sits where a blind person's hand already is: on the cane.
Read →A bedside hub reads live data from your smartwatch and answers with scent, the only sensory channel that can nudge the sleeping brain without waking it.
Read →A Milan startup skips the always-on digitizer that drains battery life in smart devices. Its H1 chip processes sound in analog first, cutting power for wake-word detection by up to 1,000x.
Read →A Waterloo team built the check nobody puts in front of an agent's write: not can it use the tool, but can it trust the evidence behind the decision.
Read →A five-person Seattle startup builds an autonomous robot that pulls weeds growing right next to the crop, the one job herbicide sprayers and tractor cultivators still cannot do.
Read →Three Argentine students turned a school project into a wildfire detection platform that fuses eight satellite feeds and beats NASA's FIRMS alerts by roughly 35 minutes on average.
Read →A tiny team built three neural models that predict a pest protein's shape, screen millions of compounds in hours, and flag when a prediction can't be trusted.
Read →A small Stuttgart team built a processor that runs neural-network math inside optical waveguides, claiming far lower energy because light barely heats up and one optical element replaces hundreds of transistors.
Read →A two-person startup goes after the real bottleneck for AI agents: not raw capability but reliability, by recording a workflow once and replaying it deterministically across any app, even legacy software with no API.
Read →The bottleneck for AI data centers isn't chips, it's a multi-year wait to plug into the grid. Voxel skips the grid entirely with off-grid solar, second-life EV batteries, and a microgrid that stays DC end to end.
Read →Most AI coding tools have a soft definition of success. Hardware doesn't: Verilog either passes synthesis or it doesn't. SigmanticAI points fine-tuned models and reinforcement learning at exactly that pass/fail oracle.
Read →Two founders, 40-gram drones, and a serious idea hiding behind a joke: chemical-free mosquito control that identifies the female by its wingbeat and runs the whole signal chain on board.
Read →Instead of photographing thousands of defective parts, Bucket generates the vision model straight from CAD with synthetic data, so quality control is ready before production starts.
Read →An all-software, AI-based ISP that works on the sensor's RAW signal before it's degraded: +75% object detection and −91% false positives in low light, all at the edge.
Read →One non-invasive clip-on sensor plus a neural net that disaggregates the load, turning a single meter reading into appliance-by-appliance insight, and cutting bills ~20%.
Read →While everyone races for bigger models, an Indian startup is squeezing a voice assistant into a feature phone: offline, a few megabytes, 22 languages. The constraint is the feature.
Read →A team that started by building better weather balloons now out-forecasts agencies that have owned the field for decades, because the balloons and the AI feed each other.
Read →Most engineering software runs on a geometry kernel older than the people using it. Aurorin is rebuilding that load-bearing layer from scratch: AI-native, for modern hardware.
Read →A London startup puts AI cameras over recycling conveyor belts, turning an opaque plant into a measurable system: proof that the best vision work happens where nobody expected it.
Read →A ten-person Toronto team turns food waste into a polyester alternative, and the clever part isn't the chemistry: it's that it runs on the polyester machines that already exist.
Read →There's no single best headless CMS — only the right fit for the project. Here's how I decide between Payload, Sanity, and Strapi.
Read →Notes on building fast, scalable web applications with Next.js and headless CMS — the decisions, trade-offs, and lessons from real client work.
Read →