About Declic
Declic is an innovative social platform backed by 500 Global that revolutionizes how people connect and discover events. The app enables users to organize, discover, and participate in activities across 38 major cities worldwide, from Tunis and Marrakech to London, Tokyo, and New York. Supporting English, French, and Arabic, Declic bridges cultural gaps and creates authentic real-world connections through shared interests in sports, arts, fitness, and social activities. Whether you're seeking salsa dancing, sports, Zumba, yoga, or just new friendships, Declic serves as the ultimate catalyst for fresh connections and meaningful relationships.
Project: LLM Event Recommender System
Fine-tuned a TinyLlama model using advanced SFT & DPO techniques to build a lightweight yet powerful LLM event recommender. Developed a sophisticated dialogue system capable of interpreting complex user requests including intention analysis, tone detection, and location preferences to provide personalized event recommendations.
Technologies Used
Python
TinyLlama
SFT
DPO
T5
Docker
Ollama
Transformers
Key Achievements
- Achieved 91.3% accuracy with sequential SFT→DPO strategy, outperforming 88.1% SFT baseline
- Created dialogue system to interpret complex user requests (intention, tone, location)
- Engineered T5-based data generation pipeline for automatic log-to-instruction conversion
- Containerized final model for production-level inference using Docker and Ollama
- Optimized model size while maintaining high performance for mobile deployment