Work

PIR Dashboard — Forecast Performance Review Platform

Redesigned Coles Group's Post-Implementation Review dashboard — adding subcategory granularity, automated RMSE pipelines, and Vega visualizations for forecasting model evaluation at enterprise scale.

Role

Data Science Intern

Year

2025

Screenshot 1
1/1
1,800+ Stores Covered
Automated RMSE Workflow
4 Stakeholder Tiers
20 weeks Placement
Power BIRSQLSnowflakeVegaSharePoint

Context

Built during a 20-week Data Science internship at Coles Group within the Data & AI division’s Command Centre. The PIR (Post-Implementation Review) dashboard is used by demand planners, supply chain analysts, and Data & AI leadership to evaluate how the Smarter Forecast AI model performs against the legacy statistical model during major retail events and promotions.

Problem

The existing dashboard had three core limitations: performance could only be reviewed at category level (no subcategory drill-down), the comments section had no backend pipeline, and RMSE comparison required analysts to manually run R scripts, download files, and type results into Excel — a process that was slow, error-prone, and often skipped.

What I Built

  • Dynamic M Queries — Converted static Power BI queries into filter-responsive ones, enabling seamless category-to-subcategory switching for deeper performance analysis
  • Automated RMSE Pipeline — Developed an R script that auto-detects the previous week’s date range, calculates RMSE for both AI and statistical models, and writes directly to Snowflake — eliminating the manual Excel workflow entirely
  • RMSE Comparison Page — Built a dual-axis Vega combo chart with RMSE on the primary axis and % performance difference on the secondary, including a time-period slider, grouping filters, and summary metric cards
  • Comments Backend Blueprint — Designed the full SharePoint-to-R-to-Snowflake pipeline for a stakeholder comments section, documented for the team to operationalize post-placement

Impact

The automated pipeline eliminated a major weekly bottleneck. The enhanced dashboard reduced manual workload, improved interpretability, and enabled evidence-based decision-making for supply chain and forecasting teams across 1,800+ retail stores.