
I Built an AI That Turns GitHub Issues Into Pull Requests — No Local Setup Required
What if fixing a bug was as simple as pasting an issue link and clicking one button?
A running log of AI systems I'm building — RAG pipelines, agentic tools, and the engineering decisions behind them — plus the occasional opinion on where this space is headed.

What if fixing a bug was as simple as pasting an issue link and clicking one button?
Building a Retrieval-Augmented Generation (RAG) system that performs well in real-world scenarios is rarely a straight path. My initial implementation started with a standard pipeline, but the evaluation results were not satisfactory. This blog walks through the iterative improvements I made—from data preparation to advanced retrieval and chunking strategies—that eventually led to a significantly better-performing system.

Diving beneath the sleek chat interfaces to find out why LLMs fail at basic counting, why 'answer only' breaks them, and the science of RLHF
