RCRitik Choprainspeedengineer.hashnode.dev·Aug 10 · 4 min readThe Bug That Made My 8-Core Box Perform Like It Only Had OneI remember staring at the throughput graph for a solid ten minutes, convinced our monitoring was broken. We'd just split a stats-collection hot path across eight worker threads. Each thread had its ow00
RCRitik Choprainspeedengineer.hashnode.dev·Aug 9 · 3 min readThe Field Nobody Read That Took Down Three Services AnywayI have a rule now, and I learned it the expensive way: never trust "nobody reads this field" as a reason to remove it. Here's what happened. We had a field in an internal API response — response.legac00
RCRitik Choprainspeedengineer.hashnode.dev·Aug 8 · 5 min readI Blamed Our Code for Dropped Video Calls. The Kernel Was Guilty All Along.Every alert dashboard was green. That's what made it maddening. Our video platform wasn't down. Nothing had crashed. But every few seconds, users would hit a frozen frame and a burst of robotic audio,00
RCRitik Choprainspeedengineer.hashnode.dev·Aug 5 · 4 min readI Doubled My Agent's Context Window. Its Accuracy Dropped 20%.I did the thing every engineer does the first time a RAG pipeline feels fragile: I assumed the fix was more context. The setup was simple. An internal support agent, retrieving from a knowledge base, 00
RCRitik Choprainspeedengineer.hashnode.dev·Aug 1 · 4 min readThe Backup That Was Already Dead When We Needed ItThe worst part of the incident wasn't the corrupted file. It was the third restore attempt. The first restore, from the previous night's backup, came back corrupted — same bad bytes, same broken check00