Lilly Chen, a former Facebook engineer and former monk, has built FSH Technologies into a Philadelphia government software company with $25 million in total funding, including a $20 million Series A assembled in about a week. The startup focuses on school food services and public records management, using AI for recognition, translation and search rather than a chatbot layer. The case matters because it shows a venture-backed path into slow public procurement without relying on hype around agents.
How FSH took shape and who backed it
FSH, pronounced "fish", took its current form in 2024 after several years of experimentation. Contrary wrote Chen her first check in 2021, before she had settled on a business idea, after an introduction through Envision, a student-run accelerator for nontraditional founders. Contrary also led the seed financing, while the Series A brought in Lachy Groom, General Catalyst, Ramp, Scribble Ventures, Operator Partners and Basis Set Ventures. The Contrary investor who first backed her, Jason Chen, has since joined the company as chief operating officer.
Before FSH, Chen left high school at 16 amid hospitalizations for Graves disease and spent about two years as a monk, ordained in the United States and later living several months at a monastery in China. She then homeschooled herself, entered Colorado College without a high school diploma, turned 20 during her freshman year and graduated in 2019 with a double major in mathematics and economics. In college she briefly played Dota 2 professionally and played against OpenAI game-playing bots, then worked in Boston and joined Facebook building infrastructure used to train AI models.
Chen said she decided she could no longer work in Big Tech while watching coverage of the Jan. 6, 2021 attack on the U. S. Capitol during a remote company meeting, questioning Facebook role in enabling organization of violence. She returned to questions from her monastic period about what made her happy, what mattered and what she was good at, with programming as her most concrete skill. She concluded she wanted to tackle everyday problems that make lives harder, even without Silicon Valley excitement, asking what it takes to solve boring work. Government software fit that search because permits, licenses and grants stay slow for reasons deeper than interface design.
What this means for buyers of government software
For public agencies, FSH offers software for K-12 food services, including meal planning, nutritional information and reimbursement requirements, alongside records and archive management that still forces residents to visit offices for paper documents. AI is applied to optical character recognition, translation and natural-language search inside the workflow, not as an agent handling an entire process. For small districts and offices, that design could reduce retrieval time and paperwork errors without retraining staff around a chatbot, while large agencies gain cleaner links between departments that currently store information separately.
Procurement remains the constraint that buyers and vendors should plan around, since FSH generated less than $100,000 in revenue in its first year and at interview time had roughly eight or nine employees with additional offers outstanding. Website updates can require approvals from policymakers, communications staff and IT teams, so adding conversational tools changes little until data ownership and procedures are mapped. Decision makers should ask how records are digitized, how reimbursement rules are maintained, and how search handles multilingual requests and legacy paper formats.
The marker to watch is whether FSH converts its fast Series A into multi-year contracts beyond early food service and records niches while keeping a team of under ten plus new hires. New district deployments, renewal rates and measurable cuts in processing time would confirm that system-first modernization beats interface fixes in government. If those contracts do not appear, the funding will look like investor patience rather than product-market fit.
