I'm Bo.
I'm a Member of Technical Staff at @perplexity_ai, working at the intersection of search and AI. I built a small research team at Perplexity Berlin. Hiring is currently paused, with rare exceptions for high-profile candidates. My current work includes:
- citras: A high-performance representation learning framework for training dense, late-interaction, and multimodal representation models.
- pplx-embed-v1: Diffusion-Pretrained Dense and Contextual Embeddings. [arXiv] [blog]
- context pruner: Query-Aware Context Compression for Better Snippets. [blog]
- q2d-web: A Large-Scale Benchmark for Retrieval in Agentic RAG Systems. [arXiv] [blog]
- pplx-embed-v2-late: Multimodal embeddings beyond a single vector. [arXiv] [blog]
- pplx-embed-v2-context: Contextual Embeddings Beyond the Gold Chunk. [arXiv] [blog]
- decider: Perplexity Decider v1.3: Training, eval, and use cases. [blog]
- search quality: Supporting the team's efforts to improve search quality across the stack.
- continued pre-training and multimodal pre-training: Mining high-quality training corpora for Perplexity.
Previously, I worked as an Engineering Manager at @JinaAI_, where I worked with my colleagues to build the best neural retrievers in the world. Our work included:
- jina-embeddings-v1: A Novel Set of High-Performance Sentence Embedding Models. [arXiv]
- jina-embeddings-v2: 8192-Token General-Purpose Text Embeddings for Long Documents. [arXiv] [blog]
- jina-embeddings-v2-bilingual: Multi-Task Contrastive Learning for 8192-Token Bilingual Text Embeddings. [arXiv] [blog]
- jina-clip-v1: Your CLIP Model Is Also Your Text Retriever. [arXiv] [blog]
- jina-colbert-v2: A General-Purpose Multilingual Late Interaction Retriever. [arXiv] [blog]
- jina-embeddings-v3: Multilingual Embeddings With Task LoRA. [arXiv] [blog]
- jina-clip-v2: Multilingual Multimodal Embeddings for Text and Images. [arXiv] [blog]
- jina-embeddings-v4: Universal Embeddings for Multimodal Multilingual Retrieval. [arXiv] [blog]
- modality gap: The What and Why of Text-Image Modality Gap in CLIP Models. [blog]
- language gap: Bridging Language Gaps in Multilingual Embeddings via Contrastive Learning. [blog]
Collaborations:
- listwise reranking: Leveraging Passage Embeddings for Efficient Listwise Reranking with Large Language Models. [arXiv]
- air-bench: Automated Heterogeneous Information Retrieval Benchmark. [arXiv]
Before joining @JinaAI_, I worked as a Machine Learning Engineer and Research Intern, and I received my MSc from @tudelft, Delft, the Netherlands, and my BSc from Lanzhou University (@iLZU1909), Lanzhou, China.
Bo Wang
November 22, 2026