Physics-based food-composition measurement layer for appliances and health platforms

Kalorie makes food composition a measured number rather than an estimate. Food is the largest daily input to metabolic health and one of the last still guessed at: sleep, glucose, and biomarkers are instrumented, but what people eat is estimated from photos or databases, with error often exceeding 30 percent on a real mixed meal. Laboratory analysis costs $200 to $800 per sample and takes days.

Our solution is a measurement layer combining multimodal sensing with a physics-constrained model library, designed to measure calories, protein, fat, and carbohydrates from a real plate in seconds. Rather than pattern-matching images, we solve an inverse problem across sensor modalities, an approach intended to generalize to real-world meals difficult for camera-only methods. The library runs on any subset of the sensor stack, allowing partners to choose lower-cost or fuller configurations and enabling us to quantify the accuracy/cost tradeoff as sensing modalities are added.

We license this layer to appliance makers and health platforms rather than building a device, closer to ARM or Dolby than a hardware company.

Proof to date: a signed calibration and evaluation program with an ISO/IEC 17025 accredited laboratory covering 1,130 analyses; a signed engineering partner building the sensor bench and reference prototype; a filed provisional patent on the subset-operable architecture; and a USDA AFRI grant submitted with the University of Arkansas. Our design target is calorie error of 10 percent or less on unseen mixed meals, evaluated against ISO/IEC 17025 ground truth.

A proof-of-concept engagement, over three to six months:

Kalorie designs and runs the experiment. We select the food set to span the composition and matrix space the model must generalize across, complete the sensor bench and reference prototype, manage laboratory analysis against ISO/IEC 17025 chemistry, fit the model library, and evaluate on held-out samples. Accuracy is measured experimentally, not assumed. The partner identifies the food categories central to its products, shares the accuracy, cost, and form-factor requirements the technology must meet, provides a technical contact, and funds the work.

We scope engagements at two levels. A category program covers the partner's categories, typically 75 to 150 samples selected to span them, and delivers measured accuracy on those foods. A platform program funds the bench build and the full 1,130-analysis calibration program, with early access to the broader model library as it generalizes beyond any single category. Budget and timeline are set with the partner.

The partner receives a measured accuracy report: the error the system achieves on its categories, benchmarked against our design target and accredited laboratory ground truth, plus a readout of which sensor configurations meet its requirements and what each costs in bill-of-materials terms.

Partner inputs, requirements, and results remain confidential. Kalorie retains ownership of the platform, generalized models, algorithms, and non-partner-specific learnings from the program. Early partners receive first-look rights on licensing in their category.

We can structure the engagement as sponsored research, co-development, or a paid pilot, and welcome partners interested in testing an early handheld unit with users.

Our founder brings 25 years commercializing sensing, human-machine systems, and deep technology, including 10 patents and a prior founder-led exit.