May 2025 – Present

AI Research Scientist

Apodex, Singapore

Pretraining data infrastructure

Since May 2026, I have worked on pretraining data acquisition.

  • Main contributor to building the pretraining data acquisition pipeline from scratch, scaling web crawling to frontier scale across web pages, PDFs, and other information-rich documents.
  • Built the large-scale, high-concurrency crawling cluster with its deduplication and database layers, and optimized its throughput.
  • Developed crawl prioritization and data quality assessment to decide what is crawled and passed on for training.

Agent harness, evaluation, and multi-agent systems

From May 2025 to May 2026, I built the deep-research agent harness from scratch and led its core architecture as development grew from a solo effort to four or five contributors. It is the harness behind the open-source MiroThinker models and the Apodex models.

  • Designed multi-source verification that checks agents' research claims against external evidence, with parallel tool use and multi-agent coordination for gathering that evidence.
  • Built the agent evaluation framework with contamination checks, benchmarking on BrowseComp, GAIA, and HLE.
  • Worked on training-data synthesis and tool development for the MiroThinker and Apodex models.

Current Public Work

  • Apodex 1.1 — Research agents that work with files, data, code, and tools. Public model: Apodex-1.1-mini. FrontierAgent harness.
  • Apodex 1.0 — Verification-focused agents for deep research. Public model: Apodex-1.0-mini.

Earlier Projects

Earlier Publications

← Back to home