Overview

Basilica was a San Francisco-based developer API company founded on December 15, 2018, and part of Y Combinator's Winter 2019 batch.^[1][2] The company built an embedding API that converted images and text into high-dimensional numerical vectors — the mathematical representations that power machine learning classifiers — allowing developers without deep ML expertise to build accurate AI models using roughly 1,000 labeled examples instead of the 1 million typically required.^[7]

Basilica was killed by platform aggregation. The core value it offered — pre-trained neural representations accessible via a simple API — was absorbed for free by Google (BERT), OpenAI, and Hugging Face before Basilica could build the domain-specific moats or network effects it promised. The competitive clock started running before the company even entered YC: Google released BERT one month before Basilica's public launch.

Basilica raised only its YC pre-seed, estimated at approximately $150K,^[16] and never disclosed a follow-on round. The company quietly wound down — no shutdown announcement, no post-mortem, no acqui-hire. Jorge Silva moved to Figma as a Senior Staff Software Engineer;^[22] Eric Fung founded Colega AI, an AI startup for small business social media marketing in Asia;^[23] Michael Lucy's post-Basilica trajectory remains unknown.

Founding Story

Basilica was founded by Michael Lucy, Jorge Silva, and Eric Fung — though the exact team composition at any given moment is slightly ambiguous. The YC company directory lists only two team members,^[3][4] while Crunchbase adds Fung as a third co-founder. Whether Fung was a co-founder from day one or joined and departed early is not documented in any public source.

The founders came from engineering backgrounds at recognizable institutions — Lucy from the University of Chicago,^[4] Silva from the University of Puerto Rico,^[5] and Fung from Penn State.^[6] Silva had prior engineering experience at MuleSoft, Runnable, and RethinkDB before Basilica — a background in developer infrastructure that directly informed the product's API-first design.^[21] No prior ML research credentials are documented for any of the three founders, which is relevant context: Basilica was not a research lab spinning out novel model architectures. It was an engineering team building accessible infrastructure on top of existing deep learning advances.

The founding insight was practical and well-timed — at least in theory. By late 2018, pre-trained neural networks had demonstrated that the representations learned on large datasets (ImageNet, large text corpora) transferred remarkably well to new tasks with small amounts of labeled data. This "transfer learning" capability was real and powerful, but accessing it required ML expertise that most software developers lacked: setting up GPU infrastructure, selecting and loading model weights, writing inference pipelines. Basilica's bet was that there was a large market of developers who wanted the output of transfer learning — accurate classifiers from small datasets — without the infrastructure overhead.

The team was building in public before they even entered YC. On November 19, 2018 — nearly four months before Demo Day — they posted a "Show HN" on Hacker News, publicly launching the API and inviting developer feedback.^[26] The stated vision was ambitious: "word2vec for anything" — a universal embedding layer for every data modality, starting with images and text and expanding from there.^[8] The founders described their goal as making the power of deep learning accessible to any developer, regardless of ML background.

The early public launch was a double-edged move. It demonstrated genuine builder instincts and a desire for real-world validation over stealth development. But it also exposed the product to immediate competitive scrutiny at the worst possible moment: Google had released BERT just weeks earlier, and the Hacker News community was quick to point out the implications. The feedback the founders received on that November 2018 thread was, in effect, a preview of the market dynamics that would eventually end the company.