Cloud Monitoring Platform
Architected a real-time infrastructure monitoring platform ingesting 2M+ metrics per minute with automated incident response and alerting.
A fintech scale-up was flying blind between deploys — incidents found by customers, post-mortems built on guesswork, infrastructure spend outrunning traffic. We built a real-time observability platform that turned log-diving into timeline-reading and made cost a number every team could see.
A fintech scale-up was flying blind between deploys: incidents were discovered by customers, post-mortems relied on guesswork, and infrastructure spend grew faster than traffic.
A real-time monitoring platform ingesting metrics from every service, with automated anomaly alerts, incident timelines and cost attribution per service. On-call engineers went from searching logs to reading a timeline.
A high-throughput ingestion pipeline on AWS handling 2M+ metrics per minute, with downsampled long-term storage and a TypeScript query layer feeding the dashboards. Alert rules are versioned in the repo alongside the services they watch.
Key Features
Real-time metric ingestion
Over two million metrics a minute from every service, queryable within seconds of being emitted.
Automated anomaly alerts
Alert rules versioned in the repo beside the services they watch — no more dashboards that drift out of sync with reality.
Incident timelines
On-call engineers read a reconstructed timeline instead of grepping logs across a dozen services at 3 a.m.
Per-service cost attribution
Infrastructure spend broken down by service, so cost conversations start from data, not blame.
Technical Decisions
Ingestion built for throughput first
The pipeline was designed around a hard number — 2M metrics/minute — with backpressure and downsampling settled before a single dashboard was drawn.
Alert rules as code
Monitoring lives in the same repo as the services it watches, reviewed in the same pull requests — so observability evolves with the system, not behind it.
Downsampled long-term storage
Full-resolution recent data, downsampled history. Engineers get the detail they need now and the trends they need later without an unbounded storage bill.
80%
reduction in mean time to recovery
45%
infrastructure cost reduction
2M+
metrics ingested per minute
Incidents became minutes-long events caught internally instead of customer-reported outages, and the cost attribution view paid for the project within the first quarter.
Development Process
Discovery
Sitting with the on-call rotation to learn how incidents actually unfolded — and where the existing tooling went silent.
Ingestion foundation
The high-throughput pipeline built and load-tested to its target before any UI work began.
Dashboards as product
Internal dashboards designed with product-level care, because a tool engineers avoid is a tool that failed.
Cost layer
Attribution added once the metric foundation was trusted, turning the platform from reactive to strategic.
Future Roadmap
- Predictive incident detection
- SLO budgeting dashboards
- Automated cost anomaly alerts
- Self-serve service onboarding