The useful question arrived before the useful object. In 2017, Madhav Lavakare was 16 and trying to understand why a classmate could not stay inside an ordinary classroom conversation. Text could appear on a phone, but the phone forced an ugly choice: read the words or watch the person saying them. A face carries timing. Hands carry emphasis. The room carries context. Looking down captured the transcript and misplaced the conversation.
Lavakare’s answer was almost childishly direct: put the captions where the eyes already wanted to be. The difficulty was that he knew little about optics. So he reached for an available teacher, the family movie projector, and took it apart. There is a fine line between research equipment and an appliance your parents would quite like returned to the shelf. Lavakare crossed it with a screwdriver.
Within months, he had a crude prototype. It was the first of many. The early device, later called TranscribeGlass, streamed text to a small display in front of the wearer’s eye. The central idea survived every revision: help someone read what is being said while still looking at the speaker. It sounds simple in retrospect, as good interfaces often do. The first customer would wait eight years.
“Why can’t we take closed captions and put them in your field of vision?”Madhav Lavakare, describing the original insight
An apprenticeship in making things
The projector episode was not an isolated household offense. At six, in Palo Alto, Lavakare built a solar oven from aluminum foil after his mother declined to let him use the kitchen oven. At nine, he made a burglar alarm for his bedroom and later admitted that it was annoyingly difficult to disarm. By 12, he had assembled a Raspberry Pi voice assistant. At 13, he was writing applications for Google Glass.
These details tempt the tidy mythology of the born inventor. The more interesting pattern is less decorative. He repeatedly found something he did not understand, opened it, and tried to make it answer a human need. Childhood tinkering became a method: build, watch what happens, listen to the person using it, then build again.
His first caption display looked like evidence from a particularly ambitious science fair: exposed components, a small transparent surface and more optimism than industrial design. Its virtue was not elegance. It made the question testable. Could a wearer read text and keep another person in view at the same time? Once the answer was yes, every ungainly component became a candidate for reduction.
This is where Lavakare’s early instinct for affordability met the merciless physics of the face. A desktop can hide weight under a table. Glasses place every gram on the nose and ears. Larger batteries buy time and impose bulk. Brighter displays improve legibility and consume power. Each improvement sends an invoice to another part of the system. The work was less a grand leap than a long negotiation among light, heat, focus, latency, price and the simple wish to forget that a device is being worn.
That method mattered because a heads-up display is a collection of small arguments. It must be readable without blocking the world, light enough to tolerate, quick enough that captions do not trail the speaker, and ordinary enough that wearing it does not turn every conversation into a product demonstration. Speech arriving from a clean laptop microphone is one problem. Speech arriving across a restaurant, over chairs scraping and three adjacent conversations, is another creature entirely.
The conversation stays in one line of sight
The PowerPoint intermission
When Yale admitted Lavakare in 2019, he did not simply ask his parents to delay the expensive and respectable plan. He presented a 15-minute PowerPoint arguing for a gap year. One gap year became two. Both went into the glasses.
The pause gave him time to join an incubator program at the Indian Institute of Technology Delhi, seek grants, meet people at technology conferences and continue testing. He worked with caption users on weekends and brought successive versions to hundreds of people. Each session narrowed the distance between an inventor’s idea of convenience and a wearer’s experience of it.
The project acquired recognition early. In 2018, Lavakare filed a provisional Indian patent for an affordable assistive smart glass. In 2019, he received the Pradhan Mantri Rashtriya Bal Puraskar for innovation, along with a presidential gold medal. Awards made the project easier to notice. They did not make the optics lighter, the battery last longer or the captions arrive sooner. Hardware remains unmoved by applause.
Lavakare eventually arrived in New Haven and studied computer science at Yale. He kept the company alive during breaks and summers, collecting a reported $70,000 across pitch competitions at Cornell, Tulane and Yale. His senior thesis focused on extracting clean speech transcripts from noisy, multi-speaker audio. The teenage hardware problem was becoming an academic specialty.
In summer 2024, that specialty took him to the MIT Media Lab as a visiting student researcher. He worked on audio AI and contributed to a live musical performance built around human and artificial intelligence collaboration. It was a different stage, but it demanded the same attention to live sound: separate signal from commotion, react quickly, and avoid getting in the human’s way.
First caption-glasses prototype assembled in Delhi.
A Yale admission, a 15-minute family presentation and the first of two gap years.
An early demonstration travels across social platforms and demand outruns inventory.
Audio-AI research at MIT; AirCaps founded with Nirbhay Narang.
The first pair reaches a customer, eight years after prototype one.
AirCaps joins Y Combinator in San Francisco.
When attention arrives before inventory
A video of an early device began spreading online in 2023. Reposts multiplied. Lavakare later put the audience at 75 million views, and more than 10,000 people joined a waiting list. The messages were often intimate: people wanted to join family dinners, follow a parent, or remain part of the rapid give-and-take around a table. The internet had delivered the flattering half of product-market fit. People wanted the thing. The other half was awkward: there was not yet enough thing to sell.
Virality is a bright, brief weather system. Manufacturing is geology. By summer 2024, Lavakare had selected Vuzix in Rochester, New York, to customize glasses for the company. The project was rebranded from TranscribeGlass to AirCaps, and its scope widened. Captions remained the wedge. Translation and meeting assistance pointed toward a larger idea: an interface for in-person conversation, available at the moment a phone or laptop would interrupt it.
“It was a thousand times harder than I could ever have imagined.”Lavakare on the eight-year build
The first customer pair shipped in late February 2025. Lavakare graduated from Yale that May. In the fall, AirCaps entered Y Combinator with co-founder and CTO Nirbhay Narang, whom Lavakare had met at an augmented-reality hackathon the previous summer. Narang had also been building voice software for smart glasses since high school. Their shared obsession had the useful property of being technically specific.
At Y Combinator, the engineers found themselves learning a less romantic system: sales. AirCaps reported monthly revenue rising from $14,000 in September 2025 to $93,000 in October. The company said it achieved the growth with less than $3,000 in marketing spend. Revenue is not a complete portrait of usefulness, but it has a virtue that a viral view does not. Someone must choose to pay.
The face is part of the sentence
AirCaps now describes itself as an AI copilot for in-person conversations. Its glasses display live captions, translate more than 60 languages and can capture meeting context. The current product uses microphones, speech recognition and a private visual display. The wearer receives text without handing the center of the exchange to a screen.
This is the product insight worth stealing. A technology can perform the requested task and still damage the moment around it. Phones already transcribe. Laptops already summarize. Neither belongs between two faces at dinner, on a walk, or in the middle of a crowded room. A wearable display earns its place only when the value of looking ahead exceeds the friction of putting a computer on your face.
Lavakare arrived at that equation from the human side. He did not begin by asking what else a tiny screen could do. He asked where the words could go so a person did not have to abandon the speaker to read them. The answer became optics, then audio research, then manufacturing, then a company. Its larger ambitions still inherit the original constraint: assistance should preserve attention rather than annex it.
There is a comic imbalance to the timeline. The first useful sentence took a moment to form. The useful object took most of Lavakare’s adolescence and college years to ship. Perhaps that is why the story feels less like a tale of sudden invention and more like a profile in sustained attention. The device keeps words in view. Its founder kept the problem there for eight years.