PgBouncer: From 45K to 210K Queries per Second
A benchmark-driven look at how one PgBouncer process became a CPU bottleneck, how multiple processes scaled it, and where the next limit appeared.
man karan
Backend engineer with 5+ years building production Python services, distributed geospatial data platforms, and high-performance APIs on AWS.
Navi Mumbai, India ๐ฎ๐ณ
karan --build backend-systems --with python,postgres,aws --optimize-for scale,reliability
I build Python backend systems for geospatial data, climate risk analytics, distributed processing, and high-performance APIs.
From March 2021 to May 2026, I worked at Intensel Ltd as a Software Engineer focused on backend and data infrastructure. I owned architecture, delivery, distributed processing, and production support for a climate risk platform used by enterprise customers across 10+ countries.
Feel free to reach out and say hi on X or LinkedIn. I'm always happy to connect, collaborate, and share knowledge.
2.3B+ building records50K+ jobs/dayp99 under 200ms60% faster deploymentsA benchmark-driven look at how one PgBouncer process became a CPU bottleneck, how multiple processes scaled it, and where the next limit appeared.
A benchmark-driven look at what happens when vector compression moves from 10M scale to 100M scale.
A benchmark-driven look at how much the PostgreSQL write path matters when loading large datasets.
See the complete archive at notes(7).
Owned backend architecture and data infrastructure for a climate risk platform serving enterprise customers across 10+ countries. Took changes from ambiguous requirements through technical design, automated testing, deployment, and production support; reviewed code, mentored engineers, and investigated incidents across APIs, workers, queues, databases, and deployments.
Django FastAPI Axum REST APIs Validation Automated testing
Python SQL Rust
LLM agents Natural-language queries Geospatial analytics
System design Code review Mentoring Production debugging Monitoring
PostgreSQL PostGIS Redis DuckDB AWS SQS Dask
AWS Docker Kubernetes Amazon EKS EC2 S3 RDS Linux CI/CD CloudWatch
Owned backend architecture and delivery for an enterprise climate risk platform serving customers across 10+ countries, from requirements and technical design through rollout and production support.
Built and operated near-real-time ingestion of half-hourly precipitation GeoTIFFs with a 10-minute availability buffer and Amazon S3 storage for insurance and financial-services risk monitoring and alerts.
Built an LLM-powered agent that converted natural-language requests into structured geospatial and analytics queries, using DuckDB for vector-data analysis and integrating with backend APIs and PostgreSQL/PostGIS.
Built ingestion, validation, indexing, and query workflows for 2.3B+ building records across 1.8 TB of PostGIS data, balancing batch throughput with low-latency access.
Designed authenticated backend delivery for 5.3 TB of geospatial raster data, with tuned storage access, caching, logging, and monitoring for reliable production delivery.
Built an asynchronous HTTP API for Redis in Rust with Axum and Tokio, using connection pooling and explicit backend resource management.
Open to high-impact collaborations, consulting, and OSS. Drop me a line at karan@jadhav.dev.