Systems architect. I design and operate distributed platforms that process 200B+ data points/month across 750K industrial sensors.

I believe in building from first principles — understanding the machine before writing the code. When Celery’s ETA scheduler had an unbounded memory leak, I didn’t file an issue. I traced it through Kombu’s QoS layer, found the missing prefetch cap, and shipped the fix upstream. It runs in production everywhere now.

Currently building:

StreamGate — Lock-free observability governance proxy in Go. 200K+ events/sec, zero-allocation hot path. Filters noise and redacts PII from observability pipelines before they hit your vendor. Recognized by Datadog engineering.

Cartograph — AI-powered code flow explorer. Parses codebases, discovers entry points, traces execution flows as interactive DAGs. Don’t read the code — read the story. Tested against a production Django/Celery codebase (509 entry points discovered).

Open source contributions:

• Celery — Fixed unbounded memory leak in ETA task scheduling. Shipped in v5.6.0. PR #9853
• Kombu — Added max_prefetch to QoS class, capping heap growth. Shipped in v5.6.0. PR #2348
• Celery — Made worker concurrency cgroup-aware with --concurrency=auto. Sizes the prefork pool from the CFS quota and CPU affinity mask instead of the host CPU count, so containers stop forking one process per host core. Shipping in v5.7. PR #10328

Celery is downloaded 10M+ times/month. The leak fix affects every deployment using ETA/countdown tasks; the concurrency change every containerized worker.

What I work with:

Go Python Distributed Systems PostgreSQL TimescaleDB Celery Redis RabbitMQ Kubernetes AWS OpenTelemetry

Writing: musashiforge.hashnode.dev

LinkedIn: sandeep-kesarwani