Truffle Wants to Give the Restaurant Kitchen a Brain

Walk into any restaurant at 11pm and the show out front looks effortless. Behind the swinging door, it rarely is. Someone is standing in a walk-in cooler with a clipboard, squinting at cases of tomatoes, guessing how many the weekend will eat. Someone else is matching a stack of supplier invoices against what actually showed up on the truck. The meal is art. The back office is a spreadsheet held together with tribal knowledge. Truffle - the Y Combinator Summer 2026 company that now operates as Marble - is betting that this backstage chaos is really a data problem, and that data problems have answers.

The pitch is direct. Restaurants, cafes and bars run on a patchwork of disconnected point solutions, spreadsheets, clipboards and habit. That patchwork costs the industry billions each year in stockouts, food waste and lost profit. Truffle's plan is to replace the whole patchwork with one system - an AI-native operating system aimed squarely at the part of the business the customer never sees.

01 / WHAT IT DOESCounting, forecasting, then acting

Truffle's product breaks into three moves. First, it sees. Computer vision and voice AI capture real-time inventory levels, which the company says runs roughly 90% faster than a person counting by hand. Second, it predicts. Machine-learning models forecast item-level demand - not "busy Friday," but how many portions of a specific dish - at a stated 97% accuracy. Third, and this is the part that separates it from a dashboard, it acts. AI agents run end-to-end workflows on their own: procurement, invoice reconciliation, scheduling, food prep and menu engineering, among others.

02 / THE CUSTOMERThe people holding the clipboard

The customers are restaurants, cafes and bars - and especially multi-unit operators, where the same counting-and-guessing routine multiplies across every location. These are businesses that already bought a point-of-sale system for the dining room and a scheduling app for the staff, and still open a spreadsheet to decide how much fish to order. Truffle is aimed at the operator who never wanted to become a data-entry clerk and does it anyway, twice a week, at closing.

03 / THE PROBLEMWhere the money quietly leaks

Food waste and stockouts are the twin taxes of a kitchen. Order too much and it rots. Order too little and the menu gets 86'd mid-service and a customer orders something else, or leaves. Neither failure shows up as a dramatic line item; both bleed margin a few dollars at a time, every day. The reason it persists is not laziness - it is that the information needed to get it right lives in too many places: a supplier portal here, a paper invoice there, last month's sales in the POS, and the head chef's memory of how the last heat wave went. Truffle's argument is that once all of that becomes structured data in one place, the guessing stops being necessary.

Figures are company-stated. Inventory time shown as relative reduction (~90% faster).

04 / DIFFERENT HOWNot another app on the pile

Most restaurant software adds one more login to a stack that already has a dozen. Truffle's bet runs the other way: replace the stack. There is a real crowd in restaurant back-of-house software - MarginEdge, Toast's xtraCHEF, Restaurant365, BlueCart, Craftable and forecasting tools like Lineup.ai all touch pieces of this. What Truffle is trying to stitch together is the full loop, from seeing what's on the shelf to placing the order to reconciling the invoice, with agents doing the work rather than a human reading a report and then doing the work. Whether the autonomy holds up in a real kitchen is the open question every vertical-AI company faces; the ambition is what marks it out.

05 / THE FOUNDERSA restaurant kid and a trading engineer

The team is two people who met the problem from opposite ends. Aakar Khanna, the CEO, grew up managing his family's multi-unit restaurant business, then covered restaurants and food-and-beverage as an analyst at Goldman Sachs, and was an early business-operations hire at AtoB (YC S20) before a stint running strategy and operations at the applied-AI company Kouper. He studied at the University of Michigan's Ross School of Business with a computer science minor. Arjun Chaliha, the CTO, was a senior engineer at Bloomberg leading AI/ML for electronic trading, is a CFA charterholder, and studied computer science at the University of Pennsylvania. Kitchens and trading floors, it turns out, rhyme: both are exercises in inventory, timing and margin under pressure.

06 / MODEL & MARKETSelling relief, not "AI"

The business is B2B software: sell the operating system to restaurants, most likely per location, and replace several existing subscriptions in the process. The stated seed funding is modest - about $130,000, tied to the Y Combinator batch - which is exactly the stage where a two-person team is still proving the product works in real service, not scaling a sales team. The larger context is that AI is landing first in industries drowning in repetitive, high-stakes busywork, and few industries qualify more than hospitality. If the winning restaurant software of the next decade is the one that removes the 4am inventory count rather than the one with the prettiest dashboard, Truffle is aiming at the right target.

07 / TEN YEARS OUTThe unglamorous bet

Truffle is not trying to make the food taste better. It is trying to make sure the kitchen never runs out of what makes the food, and never pays for what it doesn't use. That is a smaller-sounding promise than "reinventing dining," and probably a more durable one. The question that will matter in a decade is not whether AI can count a walk-in cooler - it clearly can - but whether restaurant operators trust an agent to place the order without a human nodding first. Truffle is early, small, and pointed at a real, boring, expensive problem. In software, that combination has a decent track record.