Projects
Technical Projects
LLM & GenAI Projects at Obello
As Head of AI at Obello, I lead the development of state-of-the-art generative AI models for graphic design and content creation. Our platform enables design and marketing teams to create on-brand content 10 times faster than traditional methods. Key achievements include implementing advanced prompt engineering techniques, developing custom fine-tuning pipelines for design-specific tasks, and creating robust evaluation metrics for generated content.
Crafting Resilient AI: A Hands-on Workshop on Attacks, Defenses, and Alignment
This project delves into the critical field of AI Alignment and Safety, exploring the challenges and opportunities surrounding the development of responsible, robust, and beneficial artificial intelligence. Using Keras and TensorFlow, audiences will learn firsthand how to implement the Fast Gradient Sign Method (FSGM) to launch attacks on image recognition models. Then, they will dive into practical solutions with Gemma and Shield Gemma, powerful tools designed to safeguard AI systems and ensure their ethical deployment.
AI Alignment & Safety Research - Project Pressure-Sense
Developed as part of the AI Safety Fundamentals course at BlueDot Impact, this research project investigates how AI systems understand and respond to pressure, both in terms of performance and ethical considerations. The project includes implementation of novel testing frameworks, development of pressure-resistant architectures, and comprehensive documentation of ethical implications in AI system responses under various pressure scenarios.
Gemini’s Open-Source Twin and The Rise of Small Language Model for Everyone
The world of Large Language Models (LLMs) is often perceived as a playground for tech giants, with complex models requiring vast resources. But what about making these powerful tools accessible to everyone (hint: making Large Language Models smaller, so we can call it SLM or MLM)? Enter Gemma, Google's open-source family of lightweight LLMs (for these I call them Small Language Model for their 2B version, and Medium Language Models for their 7B/9B versions), designed to bring the power of Gemini to a wider audience. Libraries/tools used: Gemma, Unsloth, LoRA, PEFT, FlashAttention.
Community Projects
ArXiv Digest Extra - Version 2
Personalized arXiv Paper Recommendations with Multiple AI Models This repository provides an enhanced daily digest for newly published arXiv papers based on your research interests, leveraging multiple AI models including OpenAI GPT, Google Gemini, and Anthropic Claude to provide relevancy ratings, detailed analysis, and topic clustering.
NeuroPurrfect AI
Established a community-focused technical blog and service platform providing affordable AI/ML consulting and educational resources. The platform features in-depth technical articles, tutorials, and case studies focused on practical applications of AI technologies. Regular contributions include implementation guides for emerging AI technologies and best practices for responsible AI development.
Beautiful Mind Vietnam
I founded Beautiful Mind Vietnam (BMVN) in 2015 and it is still currently Vietnam's first non-profit organization dedicated to mental health support, reaching over 200,000 followers across social media platforms. The organization provides free mental health services, including psychological first-aid and peer support, educational resources, and community support programs. We've also supported and even rescued more than 1000 anonymous people with mental health & abuse issues. Beautiful Mind Vietnam won 2017 WECHOICE AWARD for its outstanding social impact, and also being featured on Netflix's resource page for mental health.