Hey there, my name is Daniel,

NLP research project at UNSW: does fine-tuning LLaMA 3.1 8B on QASPER make a RAG pipeline for NLP papers ignore its retrieved context? A controlled faithfulness study across 14 fine-tuning checkpoints.

Built at the Mistral AI Worldwide Hackathon in Sydney: a voice-first health companion that runs daily check-ins by voice or phone, tracks symptoms with Mistral and writes doctor-ready PDF reports.

A MIDI Event based transformer architecture, utilizing two separate transformers operating on the global MIDI events and local event parameters.

A literature review of AI music generation: audio tokenisation with VQ-VAEs and neural codecs, autoregressive transformers, diffusion approaches, and models such as YuE, MusicGen and Jukebox.

A real-time streaming pipeline correlating German news sentiment (GDELT) with Spotify Top 200 musical features, using Kafka and Spark Streaming for dual-stream aggregation.