Projects

A selection of technical projects involving machine learning, dashboard analytics, and predictive modelling. Each card includes a short description, a skills snapshot, and a link to explore the full work.

Evaluating Shazam-Style Fingerprinting and CLAP Audio Embeddings Under Degraded Conditions

NYU Group Project · Deep Learning · 2026

Benchmarks two fundamentally different audio retrieval approaches – Shazam's deterministic landmark fingerprinting and LAION-CLAP neural embeddings – against a degraded audio dataset of 17,982 files derived from the GTZAN corpus. Tests cover additive noise (white, crowd, street at 0/10/20 dB SNR) and musical transforms (pitch shift ±1–3 semitones, lo-fi bandpass filtering). Key finding: Shazam and CLAP Music both collapse on pitch-shift, while CLAP General – a generalist model not designed for music – proves the most robust of the three.

View notebook →
Python LAION-CLAP PyTorch Librosa Scikit-learn DSP · Spectrogram Analysis Audio Augmentation Cosine Similarity Evaluation Pipelines Jupyter Notebooks
View project on GitHub →

Discogs Recommendation Model

Personal · Machine Learning · 2025

A personalised vinyl recommender built using the full Discogs catalog. It runs a five-stage pipeline: ingesting the monthly Discogs data dump, syncing my collection and wantlist via the API, generating a 50k candidate pool leveraging label family graphs and artist/style affinity, ranking by metadata score, then re-ranking the top 300 using Essentia EffNet audio embeddings sourced from YouTube to find records that genuinely sound like my digital collection. Hard caps on community haves/wants keep results underground and hopefully cheap. Check out some of my discoveries via the notebook below!

Python SQLite Discogs API ETL Pipeline Design Graph-based Affinity Scoring Essentia EffNet Audio Embeddings Cosine Similarity yt-dlp · ffmpeg Claude Code
View collection profile →
View project on GitHub →

Apple Music Listening Analytics Dashboard

Personal · Data Visualisation · 2025

A Tableau Public (i.e free tier with limited functionality) dashboard visualising my listening habits across songs, artists, albums, and genres. Includes parameters, ranking logic, calculated fields, and an integrated UX-focused layout for fun/dynamic analysis – happy clicking!

View dashboard on Tableau Public →
Tableau Calculated Fields · LODs Python ETL Time-based Filtering Dashboard UX Design Interactive Storytelling
Open full dashboard in Tableau Public →

Story time / Analysis

October 2025

Just after I moved to NYC, I started collecting my Apple Music listening data at the beginning of October, and I find it quite apt that my most listened to song for that month was 'Moving Out' by Vacations. A song about leaving home by a band from my hometown of Newcastle, Australia seems quite appropriate for that point in my life. It also looks like I mainly listened to this track on Sundays and Thursdays (Sunday listening to beat the gloomy end of the weekend homesickness followed by Thursday listening to get amped for the coming weekend in my new home?).

March 2026

This one is quite funny — my friends and I went and saw the Australian band Sticky Fingers late March at the Brooklyn Paramount. Clearly after that show all I could listen to for the rest of the month was their 2013 breakout hit 'Learn to Fly'.

April 2026

The return of Babyfather (one of Dean Blunt's many side projects) clearly had me obsessed. I inhaled 'icl' 100+ times in the last half of April alone. Whilst the song is aided by being roughly 2 minutes long, its instantly catchy beat made it hard to break the loop. April was also the month I discovered bar italia and completed a round trip to Australia in under a week, where both outbound and return flights were on Monday, meaning there was lots of spare time flying to get into some new music.

Term Deposit Subscription Prediction Model

Personal · Predictive Modelling · 2024

A supervised machine learning classifier predicting customer conversion for term deposit campaigns. Includes preprocessing, feature engineering, model training, and evaluation (ROC-AUC, CV).

Python Pandas · NumPy · Seaborn Scikit-learn Feature Engineering ROC-AUC · Cross-validation Jupyter Notebooks
View project on GitHub →