Bin Wang (Robin)

About

I'm an AI research scientist at Apodex in Singapore, where I'm the main contributor to our pretraining data acquisition pipeline. Before that, I built Apodex's deep-research agent harness and its evaluation framework.

Earlier, at A*STAR, I was a lead contributor to MERaLiON, Singapore's national audio LLM, and built the AudioBench and SeaEval benchmarks. I received my Ph.D. from USC in 2021.

papers: 60+
citations: 3,000+
h-index: 24

Recognition

Best Paper Awards at SUMEval (COLING 2025) and C3NLP (ACL 2024)
APSIPA Sadaoki Furui Prize Paper Award, 2022 and 2024
Area Chair, ACL Rolling Review (2025); Publication Chair, EMNLP 2023
Mentored 10+ junior researchers, students, and interns

Selected work

03

Pretraining data infrastructure
Apodex · 2026 — Now
Main contributor to a frontier-scale pretraining data acquisition pipeline built from scratch: a high-concurrency crawling cluster with its deduplication and database layers, plus the crawl prioritization and data quality assessment that decide what is crawled for training.
Role at Apodex
- Web crawling
- Distributed systems
- Data quality
Deep-research agents and their evaluation
Apodex · 2025 — 2026
Built the deep-research agent harness behind the open-source MiroThinker models and the Apodex models, including multi-source verification of research claims. Built the agent evaluation framework, with contamination checks, for benchmarks such as BrowseComp, GAIA, and HLE.
Apodex-1.0 report MiroThinker
- Agent harness
- Evaluation
- Multi-agent
Benchmarks for language and audio models
A*STAR · NUS · 2021 — 2025
Benchmarks and evaluation methods that others build on: AudioBench for audio LLMs (280+ citations), SeaEval for cross-lingual consistency and cultural reasoning (100+ citations), and FacEval for the faithfulness of dialogue summaries.
AudioBench SeaEval FacEval
- Benchmarks
- Audio LLMs
- Multilingual

Experience

06

2025 — Now
AI Research Scientist · Apodex
Pretraining data acquisition since May 2026. Before that, the deep-research agent harness and its evaluation framework.
2023 — 2025
Scientist · A*STAR, Institute for Infocomm Research
Lead contributor to MERaLiON AudioLLM. Built AudioBench and SeaEval, and curated 10,000 hours of Singlish speech data.
2021 — 2023
Research Fellow · National University of Singapore
Dialogue summarization and NLP evaluation: FacEval, EvalRank, and InstructDS.
2017 — 2021
Ph.D., Electrical Engineering · University of Southern California
Minor in Computer Science. Language representation learning and its evaluation, advised by Prof. C.-C. Jay Kuo.
2020
Research Intern · JD.com AI Research, Mountain View
InductivE for commonsense knowledge graph completion.
2013 — 2017
B.Eng., Electronic Information Engineering · UESTC
Undergraduate studies in Chengdu, with an exchange semester at City University of Hong Kong.