Hero

Design Lead | 2020 | Mapillary

Scaling massive geospatial datasets across tools, users, and platforms

Mapillary hosts billions of street-level images and uses computer vision to extract geospatial data for use cases ranging from OSM editing to AV training. As data volume grew, high-density regions became visually unusable: overlapping points, inconsistent sequences, and heavy load times limited productivity across the ecosystem.

I was asked to define a strategic approach to improving data legibility and access across web, desktop, and partner platforms — without removing data or breaking existing workflows.

Hero

The Challenge

Key constraints shaped the problem:

This was not a UI redesign — it was a systems-level problem where UX, data structures, and machine-learning constraints intersected.

Understanding Users & Workflows

Mapillary lacked a clear view of who used its data and how. I led a mixed-method discovery effort combining:

Outputs included a set of personas and cross-platform workflows that revealed where congestion, loading issues, and navigation breakdowns occurred.

Hero

Exploration & Rapid Prototyping

I prototyped solutions directly using Mapillary JS and Mapbox JS with real data. This allowed us to test approaches that balanced technical feasibility, ML limitations, and varied user expectations.

Three promising solution categories emerged:

Meta Sequences

Meta Sequences provided the most scalable long-term solution: create canonical paths along roads, then associate nearby imagery while letting users prioritize recency, metadata, or object detections.

This approach preserved data integrity while dramatically improving map readability — but required significant engineering investment to productionize.

Data Filtering

To deliver immediate value, I designed and tested filtering controls compatible with Mapillary Web, the Mapillary API, and JOSM’s extremely constrained plugin UI.

Hero

Key contributions included:

Automating Workflows

While out of scope for immediate release, I identified an opportunity to reduce cognitive load by detecting missing map data and auto-surfacing nearby evidence — effectively reversing the typical manual search workflow.

This concept seeded a new direction internally: targeted, single-purpose tools built on Mapillary’s data instead of generic map browsing.

Hero

Impact

Most importantly, the work reframed Mapillary’s core UX problem: from “too much data” to “how do we make big geospatial data usable, navigable, and intelligent across any platform.”