At Ethara AI, I built and curated training datasets for LLM post-training workflows, preparing and structuring data and identifying where better-targeted examples were needed. I evaluated model outputs against defined criteria and used Python and SQL to surface recurring failure patterns and data-quality issues.
For my customer churn prediction project, I built an end-to-end pipeline and a logistic regression classifier that achieved 89% accuracy. I validated it on held-out data and interpreted its drivers to develop retention strategies by segment.
In my customer segmentation project, I analysed 3,900 transactions using Pandas and PostgreSQL, then visualised findings in an interactive Power BI dashboard. The analysis informed recommendations for customer targeting and product mix.
For an order-to-cash analytics project, I engineered 22 features from 10,000 records across three sources and validated the resulting dataset. I tested an assumed quarter-end volume spike, reconciled records initially considered noise, and shared findings and costed recommendations in a report and dashboard.
