Skip to content
View davidguzmanp's full-sized avatar

Block or report davidguzmanp

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Please don't include any personal information such as legal names or email addresses. Maximum 100 characters, markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
davidguzmanp/README.md

David Guzman Piedrahita

🌍 Zürich, Switzerland | 🎓 MSc in AI @ University of Zürich | 💼 LLM Engineer @ UZH Institute for Computational Linguistics

About Me

Highly driven ML engineer with a passion for pushing the boundaries of AI. Currently diving deep into LLMs, NLP, and multi-agent systems. Always eager to learn and contribute to cutting-edge research and development.

🎯 Are you looking for a versatile AI engineer who can:

  • Fine-tune and optimize large language models?
  • Develop innovative NLP solutions?
  • Bridge the gap between research and practical applications?

Then I'm the right person for your role! With experience in academic research and industry projects, I bring a unique blend of theoretical knowledge and hands-on skills to tackle complex AI challenges.

🚀 Current Focus

  • Fine-tuning large-scale LLMs on Cerebras chips
  • Exploring multi-agent simulations with LLMs
  • Developing novel approaches in adversarial NLP

🛠 Tech Stack

Python | PyTorch | TensorFlow | SQL | Git | Hugging Face Transformers | NLP | Time Series Forecasting

🔍 Recent Projects

  • BeamAttack: SOTA adversarial attacks against NLP classifiers (CLEF2024 Competition Submission)

    • Published in CEUR-WS (click here)
    • Developed a novel algorithm that significantly outperforms traditional greedy search methods in generating adversarial text
    • Achieved up to 2.8x and 3.3x improvement in success rates against BiLSTM and BERT models, respectively
    • Implemented adaptive beam search techniques to balance attack effectiveness and computational efficiency
  • Distilling Chain-of-Thought Reasoning: Enhancing task-specific workflows

    • Created an innovative graph-to-text pipeline by breaking down complex tasks into manageable stages
    • Implemented knowledge distillation to incorporate chain-of-thought reasoning into smaller, more efficient models
    • Leveraged synthetic data from LLMs to improve performance in low-resource settings
    • Achieved a 9.22% increase in BLEU score and 12.31% improvement in TER
  • Metaphors Unveiled: Exploring LLMs for figurative text interpretation

    • Evaluated various LLMs (FLANT5 large, Mistral, Phi1.5B) on their ability to interpret and simplify figurative language
    • Combined sequence-to-sequence fine-tuning with advanced prompt engineering techniques
    • Boosted accuracy from 61.5% to 71.4% through optimized prompt design
    • Conducted human evaluations to provide insights beyond automated metrics

📫 Connect with Me

LinkedIn

Open to full-time opportunities in ML/AI research and development roles! If you're seeking a dedicated AI engineer who can drive innovation and deliver results, let's connect!

Popular repositories Loading

  1. Influence-Maximization-in-Twitter-as-a-Social-Network-Graph Influence-Maximization-in-Twitter-as-a-Social-Network-Graph Public

    Jupyter Notebook 2

  2. Graph-to-Text-LLM-with-dataset-augmentation Graph-to-Text-LLM-with-dataset-augmentation Public

    Jupyter Notebook 2 1

  3. Essentials Essentials Public

    Jupyter Notebook 1

  4. BachelorThesis BachelorThesis Public

    Jupyter Notebook

  5. SCGProject SCGProject Public

    Forked from VinciGit00/SCGProject

    Jupyter Notebook

  6. unibg-tcm22 unibg-tcm22 Public

    Forked from mldv/unibg-tcm22

    Materiale per le esercitazioni del corso Tecnologie Cloud e Mobile, Università degli Studi di Bergamo, 2022