The most revealing moment in online shopping begins with a photograph of a box on a doorstep. The courier has gone. The tracking page has completed its little ceremonial progress bar. Somewhere, a customer tries on a shirt, discovers a stubborn zipper or decides the color is better than it looked on a screen. The brand, after days of notifications, suddenly has nothing to say. Ilya Valmianski finds this silence peculiar.
His company, Signals, sends a check-in after delivery and keeps the exchange alive as a two-way iMessage thread. A customer can say the fit is wrong, ask for another color, send a picture or return nine days later with a fresh question. The thread remembers. In an industry fluent in acquisition funnels and discount codes, Valmianski is betting on an antique competitive advantage: paying attention after the money changes hands.
The idea carries a faint echo of the shop floor. Valmianski has described the best retail experience he remembers as a return. An associate came around the counter and said, “Let’s find something you’re going to love.” The sentence is warm without being syrupy and commercial without feeling extractive. Online, its usual replacement is a portal with radio buttons. Signals is an effort to recover the conversation, at software scale, before disappointment hardens into a refund.
“If you build a thread, treat it like a place, not a billboard.”Ilya Valmianski, on post-purchase messaging
A physicist follows the conversation
Valmianski did not set out to become an ecommerce founder. At UC San Diego, he earned a bachelor’s degree in biophysics, summa cum laude, in 2009, then a physics doctorate in 2017. His early papers move through fluorescently labeled brain cells, microscopy image segmentation, thin films and metal-insulator transitions. The nouns are wonderfully severe. The underlying habit is portable: take a noisy system, find the hidden structure and measure what changes.
After his doctorate, he changed direction. He joined Kaiser Permanente’s medical informatics work and moved into machine learning. One model segmented the discourse structure of clinical notes and, according to his 2022 résumé, ran in production across more than 100 million notes a year. He later led development of SmartTriage, a system designed to collect information before an appointment, prepare documentation and support decisions. It was trained on tens of millions of notes and later deployed across Kaiser Permanente Southern California.
Then came Curai Health, where he worked on natural-language understanding and generation, dialogue context, summarization and recommendation systems. His research record shows the same migration as his résumé: materials science gives way to language systems; instruments give way to conversations. By 2023, he was co-founding AuxHealth, whose consumer product became MDandMe. The company built a conversational AI experience and reported helping tens of thousands of users. Signals makes him a second-time founder.
- Biophysics
Graduates from UC San Diego, summa cum laude. - Physics
Completes a doctorate after research spanning nanoscale materials and instrumentation. - Production machine learning
Builds deployed language and decision-support systems at Kaiser Permanente and Curai Health. - First company
Co-founds AuxHealth and MDandMe around conversational AI. - Signals
Joins Y Combinator’s Winter 2026 batch with co-founder Alejandro Zaniolo.
Seen from far enough away, these are sharp pivots. Seen from inside the work, they are variations on context. What did the observation mean? What information is missing? What should happen next? Physics provided the discipline. Conversational software provided the medium. Retail supplied a new, surprisingly chatty laboratory.
The little window after delivery
Signals began as Return Signals, with a narrow argument about the timing of service. Most support starts when the customer asks for it. Most return software starts when the decision to return has already been made. Valmianski and co-founder Alejandro Zaniolo wanted to arrive earlier: after the package is opened, while the impression is fresh and before the customer has gone hunting for a form.
Zaniolo brought direct knowledge of the trade. His family ran two apparel brands with more than 200 physical stores in the United States. The founders had both sides of the fitting-room curtain: one understood conversational systems; the other knew how stores behave when a customer walks back in. Their first product description called Signals an “AI-powered luxury concierge.” The phrasing has since become plainer. Today the company calls itself an iMessage revenue channel for direct-to-consumer brands.
The useful word is “thread.” A campaign is sent and finished. A ticket is opened and closed. A thread can pause. It can resume after the replacement shirt arrives. It can contain a support request on Monday and a gift question the following week. Signals is designed around that persistence, using the same lock-screen real estate brands already occupy with shipment notices.
This also changes the product’s economic argument. Customer support is commonly measured by how many tickets software deflects. Valmianski argues that proactive, high-touch service can influence margin and lifetime value instead. If a customer needs another size, an exchange keeps more revenue than a refund. If the conversation goes well, a matching piece can be useful rather than intrusive. The distinction depends on timing, memory and restraint. A concierge who shouts coupons is merely an ad wearing a waistcoat.
The experiment is the story
Valmianski’s public writing is thick with control groups, reply rates and measured windows. Signals says a randomized test with Quaker Marine Supply included 1,910 customers, divided evenly between a post-delivery check-in and the brand’s ordinary email experience. Fifty-eight percent of those contacted replied. Seventeen percent had a support issue resolved in the thread without escalation. Over the three-week study window, the treatment group repurchased 51 percent more often.
A later A/B test with Jordan Craig produced another vivid figure: about $30 in average incremental repeat revenue per conversation within two weeks, according to the company. Thirty-six percent of contacted customers wrote back, and the check-in group returned roughly 13 percent fewer items than the holdout. These are company-reported results from individual brand experiments, not commandments carved into retail marble. Their value lies in making the proposition testable.
The experiments have also nudged Signals beyond returns. In one customer story Valmianski shared, a shopper exchanged a shirt for a larger size. Nine days later, after the replacement arrived, the thread resumed. The shopper confirmed the fit, then asked about the same shirt in navy and a men’s large for her husband. The support channel had become a place to buy again, almost incidentally.
That discovery sharpened the company’s name and pitch. Signals now emphasizes repeat revenue and continuous customer intelligence. Valmianski has written about merchandisers who like the idea of customer insight but cannot use feedback scattered across reviews, support systems and CRM software. A live corpus of post-delivery conversations offers something different: reactions tied to products, captured while impressions are fresh, then made searchable.
“The brands that win the next few years won’t be the ones fastest at reading sales data; they’ll be the ones actually listening to their customers.”Ilya Valmianski, on merchandising and feedback
Two people, one very large inbox
Signals entered Y Combinator’s Winter 2026 batch as a team of two in San Francisco. Valmianski wrote that the company went from zero lines of code to a first client in roughly a month, then to multiple paid pilots with businesses serving millions of customers. The pace is recognizably YC. The more interesting fact is what had accumulated before the sprint: years of work on dialogue, deployed models, context and the awkward edges where language becomes action.
There is a pleasant continuity in his public traces. The scientist who published a microscopy segmentation tool keeps small Python utilities on GitHub. The researcher who wrote about injecting external knowledge into language generation now wants a customer thread to remember the last exchange. His old résumé lists hiking and blues dancing, pursuits that reward rhythm, adjustment and attention to a partner. It would be too neat to make those hobbies a theory of his company. They are simply good details in a life otherwise crowded with models and metrics.
The risk in proactive messaging is obvious. A warm shop-floor gesture can become an unwelcome buzz in a pocket. Signals’ entire case depends on earning the reply, not merely achieving delivery. Valmianski’s own line about treating a thread as a place contains the standard the product must meet. Places have memory, tone and norms. Billboards have impressions.
For now, the small company is making a precise wager: online retail does not suffer from too little communication. It suffers from communication that ends at exactly the wrong moment. The receipt, shipment alert and delivery notice all concern the box. The first message about the person comes afterward.
That is where Valmianski has chosen to build. Not in the glittering instant of checkout, but in the domestic afterlife of commerce, when a sleeve is too long, a color is unexpectedly right and a customer might appreciate someone asking. The web made the store infinitely convenient. Signals would like to put the associate back in it, one open thread at a time.