Writing
Notes on backend systems, Laravel, and building things that hold up at scale.
How to Create a Custom Domain Email Without Running a Mail Server
Set up incoming forwarding and authenticated outgoing SMTP using Cloudflare Email Routing, Gmail, and SMTP2GO — eliminating VPS mail server overhead.
Why We Chose Laravel Octane for a 1M-Connection OMS
PHP-FPM couldn't hold persistent connections at exchange scale. Octane's long-running worker model changed the math.
Building laravel-tagging: Polymorphic Tags Without the N+1
How a general-purpose tagging package hits sub-100ms p99 latency at a million tags per model, and the indexing tricks that got it there.
Parsing FIX and ITCH in the Same Order Pipeline
FIX is session-based and human-readable; ITCH is a raw binary multicast feed. Here's how we normalized both into one order lifecycle.
Auto-Translating Laravel Language Files with Gemini
Why I built laravel-ai-translator - generate missing translation strings once via the Gemini API, cache them to disk, and never call the API again.
pgvector in Production: Lessons from a Real Search Feature
Adding embedding-based search to an existing PostgreSQL database instead of standing up a dedicated vector store.
Redis as a Message Bus vs RabbitMQ: When to Use Which
Both can move messages between services. They fail very differently when something goes wrong.
Debugging a WebSocket Memory Leak in a Laravel Octane Worker
Memory climbed steadily over 6 hours until the worker OOM'd. The culprit wasn't the sockets - it was an event listener.
Notes on MQTT for a Low-Bandwidth IoT Integration
Integrating field sensors over unreliable, metered cellular links meant designing for disconnection as the default state. Here is how we tuned MQTT QoS, keep-alive timers, and binary payloads.
What Building an XMPP Bridge Taught Me About Presence
"Online" and "offline" turned out to be a much fuzzier line than the protocol suggests once real networks get involved.
OCR Pipelines: Tesseract vs Cloud Vision for Bangla Text
Open-source OCR is free and private but struggled with complex Bangla script in ways hosted models didn't. Here is how we built a hybrid fallback pipeline.
Zero-Downtime Deploys with GitHub Actions and Nginx
A straightforward blue-green rollout on a single VPS, without needing Kubernetes to get there.
Queues at Scale: What Actually Breaks First
It's rarely throughput. It's almost always a slow job holding a worker hostage while the queue behind it backs up.
Docker Compose for Local Parity Across a Java, PHP, and Node Stack
Three services in three languages, one docker-compose.yml, and the small decisions that kept local dev matching production.