I am Jean de Dieu Nyandwi. I work on machine learning and multimodal AI, and currently a visiting researcher at CMU LTI, where I am fortune to work with Prof. Graham Neubig and other amazing people in Neulab and CMU.

I recently completed MS Engineering AI at Carnegie Mellon University. I did undergrad in Electronics and Telecommunication Engineering at University of Rwanda, learning machine learning on the side. Prior to that, I achieved the top score, country wide, in Advanced Level National Examinations in Electronics and Telecommunication, senior highschool.

Beside research, I also have interests in AI education and accessibility, and I spend a fair amount of time designing, writing deep dives, exploring, and sharing learning resources.

Outside technical works, I regularly workout, mostly resilience training, crossfit, calisthenics, basketball, and running.

Research Interests

I am broadly interested in machine learning and multimodal AI. More specifically, I work on improving and evaluating vision language models. I am currently exploring multimodal post-training methods, improving visual reasoning in VLMs, and making AI useful for general population.

Publications

publications = FileAttachment("pub.json").json()

// Function to format authors - handles both string and array
function formatAuthors(authors) {
  if (typeof authors === 'string') return authors;
  if (Array.isArray(authors)) return authors.join(", ");
  return "";
}

// Create the HTML for publications
html`<div class="pub-container">
  ${publications.publications.map(pub => {
    return `
      <div class="pub-card">
        <div class="pub-thumbnail">
          <img src="${pub.thumbnail}" alt="${pub.title} thumbnail">
        </div>
        <div class="pub-content">
          <div class="pub-header">
            <h2 class="pub-title">${pub.title}</h2>
            <span class="pub-year">${pub.year}</span>
          </div>
          <div class="pub-authors">
            ${formatAuthors(pub.authors)}
          </div>
          <div class="pub-venue">
            <strong>${pub.venue}</strong>, ${pub.date}
            ${pub.media ? `• As seen on: <a href="${pub.media.url}" style="color: #0366d6;">${pub.media.name}</a>` : ''}
          </div>
          <div class="pub-links">
            ${pub.links.map(link => 
              `<a href="${link.url}" class="pub-link">${link.label}</a>`
            ).join('')}
          </div>
        </div>
      </div>
    `
  }).join('')}
</div>`

Latest News

• Jan 2025: Pangea is accepted to ICLR 2025!!
• Dec 2024 - Joining Neulab at LTI CMU as a visiting researcher, working with and learning a lot from Graham Neubig and the team 🤍🥳!!
• Oct 2024 - We are releasing Pangea-7B, a fully open multilingual multimodal LLM that outperform existing open models in multilingual & culturally diverse contexts 🔥🔥. Release include models, code, training data, benchmark!!
• Oct 2024 - NaturalBench paper is now publicly available on ArXiv!!
• Sep 2024 - NaturalBench was accepted to NeurIPS 2024 - Datasets and Benchmarks Track 🥳🔥
• May 2024 - Graduated from CMU MS Engineering AI!!
• Jul 2023 - Published Transformer Blueprint article!
• Oct 2022 - Listed in top 50 AI influencers by Onalytica
• Sep 2022 - Nominated in DeepLearning.AI Event Ambassadors Spotlight 2022
• Aug 2022 - Started MS AI at Carnegie Mellon University Africa
• May 2022 - Complete Machine Learning Package is now available on web

Talks

• June 2024 - Presented Multimodal LLMs at IndabaX Rwanda. Find slides here.
• May 2024 - Talked about Vision Language Models/Multimodal LLMs in CMU 18661 Introduction to Machine Learning in Student Research Lecture.