I'm Bo.

I'm a Member of Technical Staff at @perplexity_ai, working on the intersection between search and AI. I built a small research team at Perplexity Berlin, focusing on information retrieval, representation learning, multilinguality, multimodality, and evaluation.

Some of our recent work includes pplx-embed, state-of-the-art embedding models for web-scale retrieval, and query-aware context compression for better snippets. Beyond research, I work across different parts of the stack to optimize search quality.

Previously, I worked as an Engineering Manager at @JinaAI_, where my team aimed to build the best neural retrievers in the world.

Our journey started with fine-tuning embedding models using finetuner.jina.ai. We then expanded to training models from scratch, and luckily, we witnessed the rise of RAG.

Over those years, we released general text embedding models: jina-embeddings-v1, v2, v3, v4; late interaction retrievers: jina-colbert-v1, v2; and multimodal embedding models: jina-clip-v1, v2. These models were downloaded over 50 million times on Huggingface. Although speed is key in today's AI landscape, we periodically submitted our work to academic conferences and workshops.

Before joining @JinaAI_, I worked as a Machine Learning Engineer and Research Intern, and I received my MSc from TU Delft, Delft, the Netherlands, and my BSc from Lanzhou University, Lanzhou, China.

Bo Wang

June 2, 2026