2025.5 - Present LLM post-training for AI characters

Drive e2e post-training improvements including algorithm (SFT/DPO/RL) optimization, data curation, generative reasoning reward model etc

2022.3 - 2025.5 Generative recommendation

Trillion-Parameter foundation generative model https://arxiv.org/abs/2402.17152

Non-dot-product neural retrieval https://arxiv.org/abs/2306.04039

2020.4 - 2022.3 Ads ranking models