AI Research Scientist @ Apodex · Pretraining data, agents & evaluation

Based in Singapore. I work on what goes into AI models and how to tell if they're any good: training data pipelines, agent execution, and evaluation. My earlier research spans audio-language models, sentence representations, and dialogue summarization.

• Pretraining data infrastructure — Acquisition and quality assessment of web data for pretraining, built from scratch.
• Apodex 1.1 — Research agents that work with files, data, code, and tools. Open model: Apodex-1.1-mini.
• Apodex 1.0 — Verification-focused agents for deep research. Open model: Apodex-1.0-mini.

• AudioBench — I led this benchmark for audio-language models, covering speech, audio-scene, and voice understanding (first author, NAACL 2025). Paper.
• audio-ai-hub — I maintain a curated collection of audio AI papers, models, benchmarks, and datasets. Browse the hub.

• MiroThinker and MiroFlow — Earlier work on deep research agents and their execution frameworks.
• SBERT-WK — Sentence representations from BERT-based word models (IEEE/ACM TASLP 2020).
• Research Archive on Hugging Face — Historical models and datasets retained for reproducibility.

Website & writing · LinkedIn · X · Hugging Face · ORCID · Email

Currently focused on industry work and not taking on new research collaborations.