Lj Miranda | A collection of notes, projects, and essays.
Profile

Hi! I'm Lj Miranda, and welcome to my website!

I'm currently a predoctoral researcher at the AllenNLP team at Ai2. In the past, I've worked as an engineer, consultant, and researcher, mostly in the fields of NLP and AI.

I'm broadly interested in data-centric approaches to building language technologies at scale. I'm happy to discuss research and collaborate, so feel free to reach out and chat!

 
 

Recent Posts

  • Aug 21, 2025 | Projects

    Introducing FilBench: An Open LLM Evaluation Suite for Filipino

    This National Language Month, I'm proud to introduce FilBench, a big step forward in Filipino NLP evaluation. This work was also accepted at EMNLP Main! Read to learn more about this project.

  • Aug 1, 2025 | Notebook

    Field Report: ACL 2025

    Here is my field report from the ACL 2025 Conference in Vienna, Austria. Overall, it was a great experience: the vibes are good and I'm happy to have met the larger NLP community!

  • Jul 20, 2025 | Notebook

    'Draw me a swordsman': Can tool-calling LLMs draw pixel art?

    Just a fun weekend experiment on model-context protocol (MCP): I asked several tool-calling LLMs to draw a 4-frame spritesheet of a swordsman performing a slash attack using an Aseprite MCP I built. The results were interesting!

What's New?

    Aug 2025: I’m proud to introduce FilBench, a comprehensive LLM benchmark for Filipino! Accepted at EMNLP 2025 Main. I also share some thoughts in this blog post.

    May 2025: Excited to share that I have three first & co-first author papers accepted at ACL Main: HyPER, M-RewardBench, and UD-NewsCrawl. A large collab project, SEA-VL, also got into Main!

    Nov 2024: Happy to have been part of the exciting Tülu 3 and OLMo 2 releases! My primary contribution is scaling-up our preference data using a synthetic on-policy pipeline that led to improvements in our DPO models.

    Oct 2024: Our paper on routing preference instances to human or LM annotators, Hybrid Preferences, is now available. This is the first work I co-led (with Yizhong Wang) at Ai2!

    Oct 2024: Our paper on evaluating reward models in multilingual settings, M-RewardBench, is now available. This was a fun collab with folks from Cohere for AI!

    Sep 2024: My cross-institutional collabs, Consent in Crisis and SEACrowd, were accepted to NeurIPS D&B and EMNLP 2024, respectively.