OpenChartInsight: Lightweight Automatic Interpretation of GitHub Charts

Visual analytics for GitHub repositories, including star events, activity, and contributor growth, are widely available through dashboards and plugins. Yet these visualizations often lack explanatory text, requiring users to interpret trends manually. This work presents OpenChartInsight, a lightweight AI framework that generates natural language interpretations directly beneath GitHub project charts. The framework introduces a Hierarchical Metric Interpretation method that structures repository indicators into three levels: basic metrics such as star count and forks, composite metrics such as activity and contributor retention, and trend-oriented metrics such as anomalies and long-term growth. These indicators are computed through metric templates and aligned with natural language generation. The framework is implemented with lightweight models such as Flan-T5, supported by retrieval-augmented context and a self-reflection mechanism to ensure alignment with underlying data while reducing hallucinations. To assess generation quality, we design a compact evaluation pipeline that integrates Execution Accuracy for numerical consistency, FactCC for faithful alignment, SummaC for faithful consistency checking, and BLEURT-tiny for fluency assessment. Scores from these modules are normalized and aggregated into a single composite quality score, enabling systematic and reproducible evaluation without human annotation. Experiments on GitHub repository data demonstrate that OpenChartInsight produces accurate and interpretable chart explanations, providing a practical extension to existing visualization tools for open-source project analysis.