Jainam Shah
Senior Data Engineer at Amazon. Writing about AI in production, data systems, and the gap between demos and reality.
Recent Writing
Karpathy's LLM Wiki Is Brilliant. The Access Problem Almost Killed It.
I built a wiki that auto-captures every Claude session. Getting other tools to read it required treating the wiki as a server, not a folder.
I Benchmarked Naive vs. Optimized Prompts Across 10 Job Roles. Here's What Surprised Me.
I benchmarked naive vs optimized prompts across 10 roles with PromptFoo. Sales jumped 3.02→7.86. Legal hit 410s latency. Cost is not quality.
The End of Cheap Tokens
AI compute is becoming a perishable resource. From the Qwen moment to Provisioned Throughput — how AI billing is evolving like airline tickets.