Case Studies

Work we can show you, because it's ours.

Client engagements stay confidential. These systems are from our own product portfolio, built, launched, and operated end to end by Ashton Group, and they demonstrate exactly the engineering we bring to client work.

Scaling a climate-intelligence platform to 100,000+ monthly users

Our own platform · Hydrology & climate data · 2019–present

The challenge

Build a consumer product on top of some of the most demanding public data in existence: 10,000+ USGS stream gauges, 800+ snowpack telemetry stations, and NOAA weather models, all updating continuously, and make it fast, reliable, and discoverable without a marketing budget.

What we did

We engineered real-time geospatial ingestion pipelines with daily automated map tileset publishing and fallbacks so a late upstream feed never shows users a gap. Decade-deep station histories were joined to live readings to compute percent-of-normal analytics competitors don't have. Growth came entirely from SEO architecture: thousands of location pages built from the data itself, performance-tuned to rank.

The outcome

The platform has operated continuously since 2019 and serves 100,000+ monthly active users, with growth driven entirely by organic search: zero paid marketing over the platform's life.

100K+Monthly active users
10,000+Live data stations integrated
$0Paid marketing spend
7+ yrsContinuous operation

One backend, eight production apps, zero cross-impact

Our own infrastructure · Multi-tenant platform engineering

The challenge

Operate a growing family of consumer applications, each with its own API surface, on shared infrastructure, without a deploy for one product ever breaking another, and without a team of twenty to babysit it.

What we did

Every change ships through staged, independently revertible runbooks: read-only reconnaissance, blast-radius analysis, timestamped backups, then endpoint-level verification and a regression sweep across every sibling application after deploy. The discipline is boring by design; boring is what zero-regression records are made of.

The outcome

Eight production applications share the platform today. Deploys are routine, verified, and reversible, and the operating model scales without scaling headcount.

8Apps on one shared backend
100%Deploys with verified rollback path
24/7Live operation

A production AI gateway with cost controls and failover

Our own platform · AI infrastructure

The challenge

Add AI features across a fleet of products without inheriting the standard failure modes: provider outages becoming product outages, and API bills that surprise everyone at month-end.

What we did

We built a centralized gateway routing all LLM traffic: intelligent failover across Anthropic and OpenAI, caching, retries, rate limiting, and per-feature, per-session cost attribution with hard caps. On top of it, an orchestration framework (LangGraph and FastAPI) runs multi-agent workflows against more than twenty internal tools with persistent memory and concurrency-safe execution.

The outcome

Provider incidents degrade gracefully instead of taking features down, every AI feature has a known cost per use, and new AI capabilities ship against the gateway in days rather than weeks.

20+Internal tools orchestrated
2Providers with automatic failover
Per-featureCost attribution and caps

Privacy-first computer vision for health, where video never leaves the phone

Our own product · Digital health & accessibility

The challenge

Deliver camera-guided rehabilitation coaching, pose analysis, rep counting, form feedback, for a health context where uploading video to the cloud would be both a privacy liability and an adoption killer.

What we did

All computer vision runs on-device using Apple's Vision framework: skeletal tracking, compensation detection, and range-of-motion metrics computed locally at interactive frame rates, with spoken cues and an accessibility-first interface designed for users with aphasia. Only derived metrics ever sync; the video does not exist anywhere but the user's phone.

The outcome

A clinical-grade privacy posture achieved through architecture rather than policy: there is no health video to breach, and the privacy story doubles as the product's most persuasive feature.

100%Vision processing on-device
0Videos stored in any cloud
60fpsInteractive analysis rate

Your project could be the next one here.

Tell us about your project and we'll reply with a candid assessment within one business day.

Email hello@ashtongroup.org