Hi I’m Carlos!

Co-founded Malted AI, distilling LLM solutions for enterprise. Previously at Meta AI Research, doing synthetic data for LLMs alongside the Llama 2 and CodeLlama teams. NLP Ph.D. at the University of Glasgow, on tool reasoning for LLMs under Jeff Dalton.

I led the winning team for the 2021-2022 Amazon Alexa Prize TaskBot challenge.

I explore how language models can decouple memorisation from reasoning. Specifically how they can delegate computation and knowledge by using tools. Similar to how we use calculators to do math.

I apply my research in the context of Question Answering and Conversational Systems to make language model reasoning more capable and interpretable.

News

2023

• March Invited talk at the University of Cambridge Language Technology Lab
• February Guest lecture at Conversational User Interfaces at the University of Glasgow.

2022

• December Guest lecture at the Yandex School of Data Analysis.
• November Invited talks at UCL NLP group and JP Morgan on GRILLBot research. Slides
• August New paper: GRILLBot: An Assistant for Real-World Tasks is accepted at SigDial 2022.
• July Workshop talk on Question Answering in GRILLBot with Large Language Models at SCAI 2022.
• July Organised NLP workshop at USC on Neural Methods for task-oriented dialogues.
• June Team I led wins the Amazon Alexa Prize TaskBot challenge! See our GRILLBot paper.
• June New paper: CODEC: Complex Document and Entity Collection is accepted at SIGIR 2022.
• March Invited talk at Stanford NLP Seminar on Conversational Search and Language Understanding.
• Feb Invited talk at Universidad NOVA de Lisboa on Challenges in Conversational Search

2021

• June GRILL Lab proposal I led is one of 10 accepted global teams among 125+ applicants for the Amazon Alexa Prize TaskBot challenge.
• Jan-Dec Organiser of the University of Glasgow Information Retrieval seminar.

2020

• June New paper: Relevance Transformer: Generating Code Snippets with Relevance Feedback is accepted at SIGIR 2020.