In the Long Run, a virtual running application that tracks user mileage against global routes, has released a new interactive points-of-interest mapping feature. The developer built an automated data pipeline to surface historically and culturally relevant landmarks along routes ranging from Route 66 to the Iceland Ring Road. The project highlights the practical challenges of integrating artificial intelligence into curatorial tech workflows, demonstrating how traditional data engineering remains essential for subjective content curation. The pipeline began with the GeoNames dataset, selected for its extensive geographical metadata and permissive licensing. Using Python, Apache Parquet for storage, and DuckDB for querying, the team filtered out administrative boundaries and isolated categories such as parks, monuments, and natural landmarks. A population threshold and elevation filter were applied to exclude routine settlements and minor terrain features, reducing the initial thirteen million records to approximately 725,000 globally relevant locations. Geographic calculations using Shapely and Pyproj then mapped these candidates against specific route geometries, determining proximity and sequence along the runner path. To gauge notoriety and contextual relevance, the system cross-referenced locations with Wikipedia and Wikidata. The number of language editions hosting an article served as a baseline significance metric. The developer initially intended to deploy large language models for automated summarization, but this approach proved unreliable. Anthropic Haiku, used for batched rating generation, frequently hallucinated factual details, misattributed locations, and distorted geographical data. While prompt engineering and structural grounding mitigated some errors, the developer concluded that factual accuracy outweighed the stylistic improvements LLMs offered. Consequently, the AI was repurposed strictly for subjective scoring rather than content generation, acting as a weighting factor within a broader algorithmic ranking system. The project also exposed systemic biases inherent in open data ecosystems. The reliance on Wikipedia links initially skewed results toward anglophone editing patterns, causing densely populated regions to appear as simple settlement lists rather than culturally rich corridors. Per-route parameter tuning addressed these disparities, allowing custom filters for population density, geographic clustering, and cultural categories. A hybrid scoring model balanced objective wiki language counts against the AI-generated subjective ratings, ensuring a more diverse distribution of landmarks across urban and rural segments. The final output delivers a dynamic dataset that powers the application interactive map interface. Early testing reveals significant variance in landmark density depending on regional characteristics, necessitating continuous manual oversight and iterative calibration. The development process underscores a critical takeaway for contemporary software engineering: while AI accelerates data processing and subjective scoring, it cannot replace human-curated evaluation metrics or deterministic verification. The feature is now live on select routes, with the developer noting that community feedback will drive future adjustments. The project stands as a practical case study in balancing scalable automation with the nuanced requirements of cultural and geographical curation.