What Databases Knew All Along About LLM Serving
Your inference stack is burning money on solved problems. A storage hierarchy would fix the part you’re paying for twice.
Tirtha Sarker's Blog
Writing about system architecture, AI, databases, and lessons from building in Web3.
Latest posts
Why “Skip the Code, Ship the Binary” Is a Category Error
Elon Musk says AI will make programming languages obsolete by 2026. The compiler pipeline already does what he’s describing, deterministically, in milliseconds.
ERC-8004 and Agent Reputation — Part 1: The Economics
How on-chain reputation can reduce collateral requirements when paired with the right enforcement mechanisms
Why Software Engineering Isn't Engineering
The discipline we practice has more in common with urban planning and organizational psychology than bridge building.
OpenAI Serves 800M Users with One Postgres Database. Here's What Most Teams Get Wrong About Scaling.
Routing strategies, production landmines, and the 9 rules that prevent replica failures. A deep dive into Level 2 of the scaling ladder.
What Happens When Thousands of Users Hit Your Postgres at Once (And How We Survived)
Connection exhaustion, runaway queries, and the fixes that saved our public campaign
OpenAI x NVIDIA just announced a $100B partnership.
Recently I came across this announcement: https://openai.com/index/openai-nvidia-systems-partnership/
Drift Alert: Why Your Dependencies Are a Ticking Time Bomb
How unnoticed version gaps in your dependencies lead to security risks, broken builds, and mounting technical debt
How WhatsApp Built Privacy-Preserving AI (and How You Can Too)
Inside WhatsApp’s Confidential Computing Stack for AI — A Deep Dive into Zero-Trust Inference, Enclave Security, and Privacy-Preserving Architecture at Scale.
The Harsh Truth About AI Startups: Why Most Will Fail (And How to Survive)
AI is Eating Software, But Most Founders Are Missing the Point
The Hidden Cost of AI-Generated Code: What Your IDE Won't Tell You
AI-powered IDEs are changing how we code - faster development, fewer boilerplate headaches, and instant suggestions.