Lester James V. Miranda

Timeline

1. 2011–2018

   Degrees at Ateneo and Waseda

   He studies electronics and communications engineering, with a minor in philosophy, at Ateneo de Manila University from 2011 to 2016 and graduates cum laude. With a Monbukagakusho (MEXT) Japanese Government Scholarship from 2016, he earns an M.Eng. in information, production and systems engineering at Waseda University in 2018.

2. 2018

   PySwarms and a ChainerRL internship

   His Python toolkit PySwarms, for particle swarm optimization, is published in the Journal of Open Source Software in January 2018. In 2018, he also interns with the ChainerRL team at Preferred Networks, working on efficient reinforcement learning.

3. 2018–2021

   Consultant at Thinking Machines

   From 2018 to 2021, he works as a machine learning consultant at Thinking Machines Data Science in the Philippines. In 2019, he gives a DataBeers Manila talk titled Geospatial Data at Scale with Geomancer.

4. 2021–2023

   Engineering spaCy at Explosion

   He joins ExplosionAI in October 2021 as a machine learning engineer on spaCy and works on spaCy and Prodigy until 2023. In November 2021, he writes an Explosion blog post on spaCy v3's project and config systems. His calamanCy toolkit, an open-source set of spaCy-based NLP pipelines for Tagalog, appears at the NLP-OSS workshop in Singapore in December 2023.

5. 2023–2025

   Predoctoral researcher at Ai2

   From 2023 to 2025, he is a Predoctoral Young Investigator at the Allen Institute for AI (Ai2) in Seattle. He coauthors RewardBench, a benchmark for evaluating reward models, in March 2024 and the Tülu 3 post-training report in November 2024. For OLMo 2, he works on synthetic preferences for RLHF.

6. 2024–2025

   M-RewardBench and FilBench

   Through Cohere Labs' Aya Expedition, he coauthors M-RewardBench, a reward-model benchmark covering 23 typologically diverse languages, which wins the expedition's Silver Prize in 2024 and appears at ACL in July 2025. At ACL 2025, he also publishes Hybrid Preferences, a router that sends each preference annotation to humans or language models. With four colleagues, he introduces FilBench, a benchmark for LLMs in Filipino, Tagalog and Cebuano, at EMNLP in November 2025.

7. December 2025

   Tool-use data for Olmo 3

   In December 2025, he is among the coauthors of Olmo 3, a family of fully open language models at the 7B and 32B parameter scales. As a predoctoral researcher at Ai2, he works on its synthetic tool-use trajectories for supervised fine-tuning.

8. Since 2025

   A PhD at Cambridge

   Since 2025, he is a PhD student in computation, cognition and language at the University of Cambridge's Language Technology Laboratory, working with Anna Korhonen on multilingual and equitable NLP. His first PhD paper, Polyglot Teachers, studies what makes an effective multilingual teacher model for synthetic data and is accepted to EMNLP 2026. In August 2026, he starts an internship with the Thomson Reuters Labs Foundational Research Team.