about me
- data scientist @ meta - business ai & api products
- prev @ tiktok (social) and @ quantium (government & product analytics)
- ml, deep learning & optimisation @ mit, maths + stats @ unimelb
- grew up in perth, australia 🇦🇺
recent
work experience
causal inference, experimentation, and ml modelling across whatsapp api calling, whatsapp api voice messaging, and messenger calling.
machine learning and experimentation on the social team, driving the launch and growth of social avatar, and defining social connection scenarios to improve organic relationship formation between users.
end-to-end analytics pipelines, forecasting models, and dashboards for B2B SaaS and government clients in australia.
projects
nlp & llms
fine-tuned GPT-2 and GPT DaVinci on the Shakescleare dataset to reproduce Shakespearean prose. benchmarked against a Style Transformer using BLEU, ROUGE, and cosine similarity.
built an end-to-end analytics tool that scrapes Telegram data, applies NER and TF-IDF to extract features, clusters messages via embeddings, and generates market insights using GPT-4.
machine learning
a fun side project with a friend - a travel planning app where you swipe through activities tinder-style to build a trip. i worked on data scraping, activity recommendations, and ranking. shipped to 100 real users.
built a multi-class classifier on cardiotocogram (CTG) data - fetal heart rate and uterine contraction signals - to classify fetal health states and support early clinical intervention.
predicted customer segments (A/B/C/D) for 2,627 new-market prospects using imputation techniques to handle missing data, enabling targeted sales outreach.
education
mit· masters - business analytics
university of melbourne· bsc - mathematics & statistics
university college london· exchange program - statisticsvincent.tian72@gmail.com · san francisco bay area