Alex Wiltschko - Giving Computers A Sense Of Smell - [Invest Like the Best, EP.415]

Episode Summary

Executive Summary: Patrick O’Shaughnessy interviews Alex Wiltschko, founder/CEO of Osmo, about digitizing smell with AI and chemistry. Osmo can read, write, and analyze scent to create custom fragrances, detect fakes, and eventually support health and safety applications.

Main Topics: Digitizing smell with AI (Priority: 10/5): Osmo is building software and lab systems to read, write, and model scent. Read/write loop for scent (Priority: 9/5): The company treats sensing and recreating smells as a data engine for training models. Generation: AI fragrance house (Priority: 9/5): Osmo's first commercial product speeds up custom fragrance creation for brands and creators. Counterfeit and provenance detection (Priority: 8/5): Osmo's scent sensors can distinguish real from fake products like sneakers. Scientific and personal origin story (Priority: 7/5): Alex traces the idea to childhood fascination with perfume, neuroscience, and AI. Platform and future applications (Priority: 8/5): Osmo aims to expand into wellness, safety, and broader smell-based computing.

Key Arguments: Smell can be digitized by combining molecular sensing, AI, and synthetic chemistry. Read and write together create a virtuous data loop for active learning. The core hardware already exists; Osmo is replacing the 'brains' with software. Fragrance creation is slow and secretive; Osmo can compress months into minutes. Scent is emotionally powerful because it routes quickly to memory and emotion centers. Once smell is mastered in fruits/flowers, adjacent uses like health and counterfeit detection improve.

Data Points: Expense reviews automated: 85% - Opening ad for Ramp Expense review accuracy: 99% - Opening ad for Ramp Company savings: 5% - Opening ad for Ramp Commercial fragrances in home products: 90% - Alex says 90% of home products have fragrance Molecular map dimension: 300-dimensional - Alex describes Osmo's scent map AI-designed molecules stored: 10,000, 20,000 molecules - He says the inner sanctum holds this many AI-designed molecules First teleported scent: fresh summer plum - Osmo's first fully digitized and reconstituted scent Custom fragrance timeline today: 12 to 18-month process - Traditional fragrance development for custom work Price per kilo: $10 per kilo - Lower end of fragrance pricing for some products Price per kilo: many hundreds of dollars per kilo - Upper range for fine fragrance pricing Consumer platform scale: 400 million monthly active users - Patrick cites ChatGPT adoption Consumer platform scale: 5% of the world's population - Patrick cites ChatGPT usage Sensor size: about the size of these two shoe boxes together - Description of the smaller scent sensor used for StockX Detection speed: within 20 seconds - StockX real-vs-fake detection Validation set size: hundreds of thousands of molecules - Alex describes the predictive validation experiment Blind test subset: 400 - They selected 400 very different molecules for testing

Pivotal Quotes: "We have literal rankings." — Alex Wiltschko: He explains that Osmo trains and ranks human sniffers for high-quality scent data "It's like a debugger for software." — Alex Wiltschko: He describes sniffing individual molecules as they come off the machine "We fully digitized the human sense." — Alex Wiltschko: He frames the technical milestone achieved when the plum could be read and reconstituted

Implications: Osmo still needs to shrink hardware and prove new markets, but if it succeeds, scent could become a programmable interface for products, spaces, and health.