How Phia Built a 300M Item Search Index in Their First Year ft. Phoebe Gates & Sophia Kianni

I sat down with Phoebe Gates and Sophia Kianni, co-founders of Phia, the AI alignment layer for commerce that just raised $35M at a $185M valuation with a team of under 20 people, more than half of whom are engineers.

We get into the technical decisions that actually moved the needle. They ditched third-party LLMs and APIs to build their own in-house search index, and the results were hard to ignore: 80% decrease in latency, a significant jump in search accuracy, and 40% increase in monetized GMV. We also talk through the SKU unification problem at scale, how personalization became their core technical moat, and why owning their own model was never really optional for what they were building.

If you're building AI products, pay attention to how they think about the MVP in the in-house pipeline. They use vibe coding to validate whether users actually care about a feature before committing real engineering resources to it. It's one of the more practical frameworks I've heard from a team moving this fast.

$35M raised, 1M users, 6,200 brand partners, and a proprietary search index over 300M items, all within their first year. The build story here is worth understanding.

Make sure to go signup for Phia here: https://phia.com/

Topics covered:
Proprietary search index at 300M item scale, moving off third-party LLMs, latency as a product problem, using vibe coding for feature validation, and scaling a technical team early.

00:00 Meet Phoebe Gates & Sophia Kianni : Building the AI Alignment Layer for Commerce
00:06 Explosive Early Traction: 1M Users, 6,200+ Brands, Founder-Led Growth
00:33 The $35M Raise & Why Infrastructure VCs Leaned In
01:30 Why Shopping Hasn’t Evolved (and the Two Things Consumers Want)
02:04 Under the Hood: Indexing 300M Items + A Unified Search Algorithm
02:35 The Merchant Pain: Rising CAC, Retargeting Limits, Wasted Spend
02:50 From Search to Personal AI Shopper: Data Moat & Personalization
03:35 Cutting Returns by 50%: Transparency + Better Matching
04:30 Tackling the Cognitive Load: Funding Priorities & Building the Team
05:30 In-House ML Wins: 80% Lower Latency, Higher Accuracy, +40% GMV
06:15 Ship Fast, Then Build Moats: MVPs, Base Models, and ‘Vibe Coding’
07:44 Origin Story at Stanford: The North Star of an AI Shopping Agent
09:04 Why Build Core Tech In-House: Demos, Feedback, Search & Personalization
10:37 ‘Should I Buy This?’: Resale Value, Girl Math, and Decision Factors
13:04 User Dinners, Roasts, and the Future of Agentic Commerce (Wrap-Up)

Tiff In Tech

Phoebe Gates and Sophia Kianni, co-founders of Phia (FIA), discuss the engineering decisions behind their AI-powered shopping assistant that reached 1 million users and 6,200 brand partners in 10 months, raising $35M at a $185M valuation. A key technical pivot was moving off third-party LLMs to build a proprietary in-house search index over 300 million items, resulting in 80% lower latency and 40% higher monetized GMV. They use general-purpose LLMs and vibe coding only for rapid MVP prototyping to validate demand before investing in custom infrastructure. Their personalization engine reduced returns by 50% by matching users to products based on style, size history, and price preferences. The team of under 20 people—majority engineers—runs biweekly user feedback dinners with power users to drive iterative product development.

Why General LLMs Aren't Enough: Inside Phia’s Engineering ft. Phoebe Gates & Sophia Kianni