Sagas. Time-lapse Social Network

Traditional social platforms prioritize feeds and followers while ignoring how places change over time. Sagas needed to process 500,000+ location-tagged media items and deliver map tiles in under 100ms worldwide – a fundamentally different data model from feed-based networks. The platform had to handle near-real-time media ingestion (photos and videos at geo-coordinates), generate timelapses automatically and serve them through a map-centric discovery UI that scales across 10,000+ geographic zones.

Enable users to effectively create and publish geo-located timelapses using their photos and videos. Develop a dynamic map-based discovery feature that allows users to explore compelling stories by location and time. Streamline the ingestion and processing of large-scale media in near real-time. Establish a strong foundation for AI-driven insights that encompasses change detection and trend analysis.

We brought deep expertise in high-load systems, real-time data pipelines and cloud infrastructure. Our architecture processes geospatial events at 20,000+ events/second through Flink pipelines while Cassandra handles the social graph and location metadata at scale. We built an AI-powered content indexing system that enables time-based exploration across geographic zones – driving 45% higher content discovery compared to chronological feeds. Object storage with CDN-backed delivery handles 5TB+ of media monthly.

The backend is developed in Java with Spring Boot, supporting APIs, user logic, and media workflows. Apache Kafka is utilized for event streaming, while Apache Flink processes real-time events, including uploads, location updates, and time-lapse generation jobs. This architecture ensures that the platform remains responsive, even during periods of heavy media load.

Sagas features a modern web interface designed with React and Next.js, emphasizing map-centric navigation and immersive media viewing. Users engage with locations through an intuitive map UI, timelines, and location-based feeds optimized for both desktop and mobile browsers.

Apache Cassandra is used to store user profiles, location metadata, and social graph data at scale. Object storage is utilized for photos, videos, and generated timelapses. Apache Pinot provides fast analytics for tracking location popularity and activity trends, and supports discovery APIs. The entire platform operates on Kubernetes with Istio for traffic management and is deployed on AWS, ensuring elasticity and global reach.

Users can upload photos or videos linked to specific coordinates, creating timelapses that illustrate how a location evolves over days, months, or years. Each timelapse creates a lasting narrative tied to a real-world location.

Content is presented through an interactive map rather than a traditional feed. Users can explore cities, landmarks, and regions to view active timelapses and community-created historical stories.

The platform handles media uploads asynchronously and at scale, ensuring smooth performance and quick timelapse generation. Public APIs facilitate future integrations with urban analytics tools, tourism platforms, and research applications.

We delivered a platform that merges geospatial data with temporal storytelling, processing 500,000+ location-tagged media items with sub-100ms map tile delivery worldwide. Our architecture enables a fundamentally new type of social experience that no feed-based network can replicate.

We built AI-powered content indexing that enables time-based exploration across 10,000+ geographic zones. Our recommendation engine drives 45% higher content discovery compared to chronological feeds – proving that location-temporal relevance is a more powerful engagement signal than recency alone.

We engineered object storage that handles 5TB+ of media monthly with CDN-backed delivery across 3 regions. Our Flink pipeline processes geospatial events at 20,000+ events/second with 99.9% uptime – the infrastructure foundation that makes a global map-based social network viable.