Viral traffic & edge resilience
API Gateway + WAF + adaptive circuit breakers when legitimate traffic spiked 22× overnight — graceful degradation, tenant quotas, and stale-while-revalidate instead of hero reboots.
Shipping touches production: APIs, integrations, internal platforms, and this site — Next.js, typed APIs, MySQL, and email & review flows you can actually run locally. Case studies cover client outcomes in depth; here you'll find code-first snapshots and GitHub entry points.
Highlights below map to stacks I ship in prod: event-driven backends, Laravel services, React surfaces, Kafka / queues, CDC, and AWS-grade edges. New write-ups cover viral-traffic gateway patterns and binlog-driven cache sync — jump in below.
Next.js App Router, static HTML sections, MySQL for messages & reviews, SMTP notifications, and an admin flow for content.
Rewritten URLs (no .html in the bar), sitemap/robots, and structured data for search.
Active GitHub profile with backend utilities, API experiments, and tooling around Laravel, Node, and infra automation. Star/follow for spikes on Kafka consumers, queue discipline, and pragmatic READMEs.
API Gateway + WAF + adaptive circuit breakers when legitimate traffic spiked 22× overnight — graceful degradation, tenant quotas, and stale-while-revalidate instead of hero reboots.
Debezium on MySQL binlog with granular cache busting — inventory finally matched the warehouse without cron hacks or support escalations.
OpenAPI-first REST contracts with generated TypeScript clients, Laravel route validators mirroring the same schema, and sandbox keys so integrations ship without slack-thread archaeology.
Correlation IDs propagated from edge → workers → DB spans; Datadog dashboards wired to SLO burn rates so on-call sees flame graphs, not grep tournaments.
Production-grade Stripe-style flows: idempotent webhook ingestion, Kafka or SQS-backed workers, DLQ replay UIs, and reconciliation tooling — patterns documented in case studies.
Composite indexing, read replicas, Redis cache-aside, and slow-query teardowns that moved dashboards from multi-second loads to sub-second p95 — stack traces and metrics in the query graveyard write-up.
Immutable history layers over noisy CRM syncs: conflict resolution, Zoho bridges, and auditor-friendly exports — summarized under CORPRISK.
Turning dead-letter chaos into operator-grade tooling: surfaced failures, standardized envelopes, retry policies, and support-facing replay — see dead letter purgatory.
Tenant-bound tokens, rotating API keys with hashed storage, scoped OAuth-style scopes on internal services, and audit hooks so support can revoke access without DB surgery.
Reusable modules for VPC-linked ECS services, RDS parameter families, secrets rotation, and staging mirrors so “works on my cluster” stops being a meme.
Template versioning, bounce/complaint webhooks feeding suppression lists, idempotent send IDs, and retries that respect provider caps — fewer surprises in customer inboxes.
Redis-backed toggles with per-tenant overrides, middleware that fails closed, and kill switches wired to dashboards so deploys can ship dark and brighten safely.
HMAC verification, clock-skew tolerant replay windows, normalized payloads into Kafka topics, and idempotency keys shared with billing consumers — one front door for providers.
Async CSV/ZIP builders with signed S3 URLs, row-level redaction by role, job progress UI, and retention jobs so legal asks don't stall engineering for a week.
WebSocket channels fed by Redis pub/sub for queue depth and connector lag; Sparkline-heavy React panels so ops sees CDC backpressure before customers smell smoke.
Horizon-separated queues for nightly reconciliation, mutex locks on overlapping schedules, and guardrails so a bad migration can't fork fifty duplicate settlement jobs.
Field-level auth, DataLoader-style batching to kill N+1s, persisted queries for stable caches, and complexity limits so one mobile screen can't accidentally DOS the cluster.
Expand/contract column moves, shadow writes, backfills in chunked jobs, and cutover flags so deploy Friday isn't synonymous with pager Saturday.
Topic naming conventions, compatibility checks before deploy, and consumer-driven contracts so a new billing field doesn't brick five downstream services at 3AM.
Device token hygiene, topic vs direct sends, collapse keys for noisy alerts, and dead-token pruning so marketing pushes don't silently rot your sender reputation.
Headless Chrome or wkhtml queues for consistent layouts, S3 artifact storage, idempotent render keys tied to ledger rows, and retries when fonts fail in CI-parity containers.
Index pipelines from OLTP via CDC or batched ETL, synonym maps, typo-tolerant queries, and index aliases for blue/green reindex without search downtime.
Sliding-window Redis counters per tenant and route class, challenge tiers for suspicious ASNs, and coordination with WAF so humans sail through during spikes.
Automated snapshots, documented restore playbooks tested quarterly, RPO/RTO numbers leadership actually understands, and chaos-adjacent exercises before audit season.
Scoped npm/Composer registries behind SSO, immutable artifact promotion, and Docker layer caches tuned so main builds stop waiting on the public registry lottery.
Pact-style consumer/provider checks on critical APIs, plus scheduled canaries that hit production read-only endpoints and Slack when latency crosses SLO for two windows straight.
Versioned protos in CI, deadlines/cancellation propagated across hops, and interceptors for auth + tracing so Ruby/Python/Laravel callers share one typed spine instead of REST soup.
Fivetran or custom EL from OLTP, staging layers in Snowflake/BigQuery, dbt tests on grain, and reverse-ETL only when product actually needs it — not “SQL for SQL's sake.”
Upload to S3, Sharp/imgproxy workers for WebP/AVIF variants, signed URLs with short TTL, and cache-keys busted by content hash so designers stop emailing “why is this blurry.”
Role-scoped resources, action modals with confirmation policies, bulk operations with background jobs, and audit columns so support actions stay traceable under compliance review.
Separate queues for interactive vs batch work, per-tenant concurrency caps, and back-pressure signals so one noisy neighbor can't starve everyone else in the cluster.
Ledger-grade rounding rules, snapshot FX rates at invoice time, jurisdiction-aware VAT/GST tables versioned by effective date, and immutable postings once periods close.
HMAC-SHA256 signatures, exponential retries with jitter, dead destinations auto-disabled, and a delivery log UI so integrators self-serve instead of pinging your Slack.
Scenario scripts against staging mirrors, traffic shaped to match peak hour curves, pass/fail gates in CI for p95 and error rate, and recorded baselines to catch regressions before press day.
Atomic tokens, accessible primitives, visual regression in CI, and generated prop tables so product and eng agree on what “done” looks like before pixels multiply across surfaces.
IAM roles over static keys, automated rotation hooks for DB passwords, break-glass procedures documented next to the Terraform module — fewer long-lived tokens in env files.
Deterministic bucketing by user ID, mutually exclusive experiments, guardrail metrics wired before ship, and kill switches when conversion tanks unexpectedly mid-week.
Chunked parsing with row-level error reports, preview mode before commit, transactional batches sized for lock time, and progress bars that tell ops something real is happening.
Slash commands behind SSO, confirmation threads for destructive ops, audit logs landing next to CORPRISK expectations, and rate limits so every intern can't restart prod by typo.
Deprecation timelines in Sunset + Deprecation headers, breaking-change playbooks, dual-write windows for mobile clients still on last winter's build.
Denormalized tables fed by domain events, rebuild jobs when projections drift, and explicit lag metrics so support stops blaming “cache ghosts.”
Postgres channels for coordinator broadcasts, payload size discipline, and fallback polling when connections flap — simpler than standing up Socket servers for every tiny alert.
Transaction vs session pooling modes tuned per workload, prepared statement pitfalls documented, and RDS max_connections budgets that leave headroom for migrations and consoles.
Region-aware defaults, script blocking tiers before marketing tags fire, consent receipts stored next to audit trails, and fewer “accept all” dark patterns your counsel regrets.
Duplicate requests to new stacks without affecting users, diff collectors comparing responses, and traffic caps so experiments don't double cloud bills overnight.
One-command stacks mirroring prod services (MySQL, Redis, Kafka UI), seeded fixtures with deterministic IDs, and Make/NPM scripts so onboarding stops at “clone & up,” not tribal wiki hunting.
Latency-based DNS or edge workers for auth checks closer to users, KV-backed session hints, and origin shields so regional outages degrade gracefully instead of globally.
Multi-processor routing with failover rules, wallet capture flows, payout batches matched against ledger rows, and nightly exception queues humans actually clear before month-end close.
OpenAPI-synced reference sites, runnable curl snippets, changelog RSS, and ownership metadata so teams know who to ping when an endpoint misbehaves at Friday 6pm your time.
Need something similar on your roadmap? Send stack, constraints, and timeline — I reply personally.