Diego Garcia-Olano

About Diego Garcia-Olano

Research Scientist at Meta Superintelligence Labs ( formerly GenAI Trust & Safety, formerly Responsible AI ). I work on safety alignment evaluation/mitigation for pre & post-trained LLMs on multi-modal privacy risks ( IP, PII, memorization, Biometric (image/video) and RAG related risks) establishing research direction, LLM coding, benchmark creation, evaluation and analysis of Omni & Reasoning models. I developed a research plan for my PhD intern on safety contrastive steering of fine tuned LLMs to efficiently and effectively reduce hallucinations (CASAL) which has been accepted at NeurIPS 2025 Mech Interp Workshop. See: in GenAI/RAI I developed Influence Function & Model dynamics based methods to improve data curation for both LLMs and Integrity Classifiers ( to prevent human exploitation/harm ) and improve downstream performance/sensitivity via embedding analysis ( Error Discovery by Clustering Influence Embeddings NeurIPS 2023 ). In RAI, I was part of Model Understanding explainability team which developed the PyTorch Captum library and work on LLM based explanation methods (EMNLP 2023)Graduated with a PhD at UT Austin in 2022 explainable Natural language Processing and Machine Learning with a focus on entity representation and multimodal learning for applications in healthcare & policy. During my PhD I have done 2 research internships with Google AI ( Mountain View \'18) and Google Cloud AI ( Seattle \'19) and one with IBM Research ( NY remote \'20 ) on topics involving learning dense entity representations for information retrieval tasks ( CoNLL 2019 ), explainability for seq2seq models and biomedical interpretable entity representation learning ( ACL-ICJNLP 2021 ). I\'ve additionally published on studying knowledge injection for multimodal vqa (The WebConf 22), learning classifiers that leverage learned prototypes for time series clinical data (ICML 19 (short), IJCAI 19 (long)), and automating learning of political networks from online media sources ( ECML 16 ).I did a fellowship at the University of Chicago Data Science for Social Good program ( 2016 ) and was as a Fulbright Specialist scholar leading a workshop and developing a tool on gender bias in judicial proceedings using NLP in Azul, Argentina (spring 2022). Prior to getting my masters from the UPC in Barcelona ( 2015) and PhD at UT Austin ( summer 2022), I was a full web stack developer working for start ups and social change organizations, and did many data visualizations for Spotify, Pitchfork, amongst others.

Experience

1. Research Scientist

   Meta

   Sep 2022 — Present

   Meta Superintelligence Labs Trust ( nee Gen AI Safety Alignment nee Meta Responsible AI - Model Understanding )

Education

• Universitat Politècnica de Catalunya

  Masters in Innovation and Research - Data Mining, Machine Learning, Statistics, Text Mining, Network Science

  2013 — 2015

• Cockrell School of Engineering, The University of Texas at Austin

  Doctor of Philosophy - PhD, Machine Learning (Decision Information Communications Engineering)

• The University of Texas at Austin

  Bachelor's degree, Computer Science, Political Science and Hispanic Studies (Spanish) with a minor in Business

  1999 — 2004

Skills

• Javascript
• Text Mining
• Social Network Analysis
• Css
• Postgresql
• Machine Learning
• Software Development
• Python
• Mysql
• Web Development
• R
• Data Analysis