La Yaung Linn Lett
BSc (Hons) Data Science & AI graduate (First Class), UWE Bristol. Seven end-to-end projects across SQL/Power BI dashboards, machine learning and geospatial analysis — seeking 2026 graduate roles in data analytics or data science.
Projects
UK Job Market Intelligence Dashboard Live
Pipeline (API → SQL → keyword extraction → dashboard) built on 1,200 real UK job postings. Found that data analyst roles want SQL and Power BI, while data scientist roles want machine learning — and advertise about 1.4x the salary (£57,306 vs £42,085 on the 613 postings with a real, non-estimated salary).
UK Retail Sales & Customer Retention Dashboard Power BI
Three-page Power BI dashboard analysing 541,909 UK retail transactions. Flagged 195 high-value customers (£471.7k in historical spend) as churn risks, producing five prioritised business recommendations — including two targeted retention actions — each with a defined success metric and illustrative revenue impact.
Customer Churn Prediction Live
Lifted churner recall from 58% to 77% on 7,043 telecom customers using SMOTE and decision-threshold tuning on an XGBoost classifier. Used XGBoost feature importance to identify payment method, internet service type and contract length as the strongest churn drivers.
Bristol Crime & House Prices — Spatial Regression Live
Raised explained variance (R²) from 0.11 to 0.74 across 182 Bristol neighbourhoods by replacing global OLS regression with Geographically Weighted Regression on 34,543 house transactions and 159,666 crime records. Crime's effect on price ranges from −22.6% to +2.6% depending on the neighbourhood, against a single −7.5% under the global model.
London Property Investment Scoring Code
Scored all 32 London boroughs on growth, safety and affordability using 30 years of HM Land Registry prices and 6 years of Met Police crime data. Found a −0.83 correlation between 2024 entry price and 10-year growth: the most expensive boroughs have been the worst investments, with 17 of 32 offering above-average growth at below-average prices.
Music Mood Classifier Group project
Five-person university project classifying 114,000 Spotify tracks into six mood categories at 68% test accuracy and 0.91 macro ROC-AUC. I was the feature engineer: dataset cleaning, the first 114→4 genre taxonomy, and the transformer expanding 15 raw audio features to 42. The final model and evaluation were teammates' work — the repo credits each contribution by name.
FIFA World Cup 2026 Prediction Engine Code
Rated 300 international teams with Elo from 49,477 historical results, converted rating gaps into scoreline probabilities with a Dixon-Coles Poisson model, then simulated all 104 tournament matches 10,000 times. Argentina wins about 23% of simulations. Covered by 77 automated tests running in CI.
Skills
Languages & Databases
Python, SQL, PostgreSQL, SQLite
Libraries & Tools
Pandas, Scikit-learn, XGBoost, LightGBM, Matplotlib, GeoPandas, Streamlit, Power BI, Git/GitHub, Excel
Techniques
Statistical analysis, hypothesis testing, data cleaning, machine learning, cross-validation, API data collection, dashboard development, data storytelling