Vibecoding an inference chip
Ankit Gupta’s coding agents accelerated FPGA LLM inference from 6 to 213 tokens per second, despite his lack of Verilog experience.
One thoughtful comment unlocks the ability to create threads.
Ankit Gupta’s coding agents accelerated FPGA LLM inference from 6 to 213 tokens per second, despite his lack of Verilog experience.
One thoughtful comment unlocks the ability to create threads.
A YC partner spent $10000 on Claude and OpenAI tokens and vibe coded Verily on an AWS FPGA to build an inference chip. I need to be more ambitious with my side projects.