Hi, I'm Yinhan Liu
Founder @ Stealth Mode and FlickBloom | Serial Entrepreneur (1 Exit) | 60k+ Citation AI Researcher
First author of RoBERTa, BART and mBART. Building the future of AI-powered social media marketing.
About Me
I’m an AI researcher and entrepreneur focused on making frontier open-source models practical at production scale. In Stealth Mode, I’m building a Neo Cloud marketplace with token-optimized model access, cost-aware routing, and flexible infrastructure—helping teams run production AI workloads with better economics and control. I also founded FlickBloom, where multimodal modeling and AI agents support modern marketing workflows.
As Founder & CTO of BirchAI, we raised $4M in funding and successfully led the company through its acquisition by Sagility. My research at Facebook AI Research produced RoBERTa, BART, and mBART - foundational models that have shaped the field of NLP with over 60,000 citations.
✨ ✨ Expertise
Experience
Co-Founder
Stealth ModeCEO & Founder
FlickBloomCo-Founder & CTO
BirchAIResearch Engineer
Facebook AI Research (FAIR)Products
Stealth Mode
A Neo Cloud marketplace providing token-optimized access to open-source AI models through production-ready APIs, flexible infrastructure, and cost-efficient model routing.
FlickBloom
A full-stack social media marketing agency driven by advanced technology leverages proprietary in-house multimodal models and AI Agents to revolutionize content creation, social media strategy, and end-to-end campaign management for B2B and B2C businesses.
ProductGrade
Platform connecting innovative startups with qualified expert reviewers to provide authentic, professional feedback that builds trust and drives product adoption.
BirchAI
Healthcare AI platform utilizing state-of-the-art Speech and NLP models for medical customer support. Successfully acquired by Sagility.
Research Contributions
60,000+ total citations across foundational NLP research
RoBERTa
📄A robustly optimized BERT pretraining approach that achieves state-of-the-art results on multiple NLP benchmarks. One of the most influential NLP models with thousands of citations.
BART
📄A denoising autoencoder for pretraining sequence-to-sequence models. BART combines bidirectional and auto-regressive transformers, achieving excellent performance on text generation tasks.
mBART
📄Multilingual denoising pre-training for neural machine translation. Extends BART to multiple languages, enabling cross-lingual transfer and improving translation quality.