Vairavan
Lakshmanan

The Wi-Fi portal codebase had accumulated significant technical debt — 580 lint errors, a cyclomatic complexity of 12, and code coverage sitting at just 68%. Every release was a gamble. Regressions were frequent, testing cycles were long, and onboarding new engineers was painful. The team was spending more time firefighting than building.

I led the architectural decision to migrate to an NX Monorepo structure and upgrade Angular to v16. The key insight was that the fragmentation wasn't just a code quality problem — it was a structural one. NX's enforced module boundaries, shared libraries, and affected-project detection would transform how the team built and tested features.

The tradeoff was real: a migration of this scale carried migration risk and required the team to temporarily slow feature delivery. I made the case to stakeholders by quantifying the cost of the status quo — regression cycles, delayed releases, and growing onboarding overhead.

Airline passengers connect to the Wi-Fi portal over satellite networks — high latency, low bandwidth, and highly variable throughput. The portal's 22MB frontend bundle and ~6s load time was unacceptable in this environment. Users were dropping off before the page even loaded. Standard web performance thinking didn't apply — we needed an architecture purpose-built for constrained networks.

I architected a multi-layer approach. First, a content-split strategy to decompose the monolithic bundle into smaller, route-specific chunks — reducing initial payload from 22MB to 5MB. Second, a Service Worker caching layer to intercept repeat requests and serve assets from cache, eliminating redundant satellite round-trips entirely.

The key tradeoff: Service Workers add complexity around cache invalidation and versioning. I designed a deterministic cache-busting strategy tied to our release pipeline to ensure passengers always received fresh content post-deployment without sacrificing cache efficiency mid-flight.

Delta's applications were generating rich metrics and logs, but observability was fragmented. SRE teams were manually writing CloudWatch queries to surface a limited set of metrics in basic CloudWatch dashboards. Simultaneously, some metrics were being shipped to third-party tools — adding cost, data silos, and operational overhead. There was no single pane of glass for cross-application observability.

Selected as part of Delta's Re-Guild cross-functional initiative (Fall 2025) — a delta-wide reliability and observability program — I researched and architected a PoC solution that would unify log and metrics ingestion into a powerful, queryable OpenSearch layer.

The core insight: rather than pushing data from the application layer, we could tap into CloudWatch Subscription Filters — intercepting the log stream non-invasively and routing it through Kinesis Data Firehose into OpenSearch. Zero application changes. Rich dashboards. Reduced third-party dependency.