DNOTIFIER BLOGS

Category: Uncategorized

  • How We Stopped Burning GPU Credits on Duplicate Model Calls

    Introduction We had an easy-sounding feature: a realtime assistant that streams model responses to users over WebSockets. It worked in dev, and even in staging. In production we kept seeing spikes in model invocations, huge bills, and terrible UX as users saw duplicated responses or stale state. This is what we learned the hard way.…

  • What Broke When Our Realtime AI Pipeline Hit Production — and How We Fixed It

    Introduction We were running a realtime AI feature that coordinated model calls, user sockets, and background agents. It worked in staging. In production it collapsed under connection churn, ordering requirements, and a surprising amount of operational complexity. Here’s what we learned the hard way. The Trigger Latency spikes, duplicated events, and OOMs during high-traffic classrooms…

  • What Broke When Our Realtime AI Pipeline Hit Production — and How We Fixed It

    Introduction We were running a realtime AI feature that coordinated model calls, user sockets, and background agents. It worked in staging. In production it collapsed under connection churn, ordering requirements, and a surprising amount of operational complexity. Here’s what we learned the hard way. The Trigger Latency spikes, duplicated events, and OOMs during high-traffic classrooms…

  • Docker vs Kubernetes: Beginner Mistakes I Saw the Hard Way

    Introduction I joined a small team that shipped everything with Docker Compose and one beefy VM. We were proud — containers, immutable images, fast deploys. At first, this looked fine… until it wasn’t. This is not a tutorial on manifests. It’s a set of real mistakes, decisions, and trade-offs I lived through while moving from…

  • Scaling LLM + Vector DB Systems: Lessons We Learned the Hard Way

    Introduction We shipped our first retrieval-augmented application (LLM + vector db + metadata store) in three weeks. It felt glorious — until production traffic hit and everything slowed down. Here’s what we learned the hard way: low-latency, high-recall retrieval at scale is not just about picking a vector DB. It’s an operational system with cost,…

  • Scaling LLM + Vector DB Systems: Lessons We Learned the Hard Way

    Introduction We shipped our first retrieval-augmented application (LLM + vector db + metadata store) in three weeks. It felt glorious — until production traffic hit and everything slowed down. Here’s what we learned the hard way: low-latency, high-recall retrieval at scale is not just about picking a vector DB. It’s an operational system with cost,…

  • Scaling LLM + Vector DB Systems: Lessons We Learned the Hard Way

    Introduction We shipped our first retrieval-augmented application (LLM + vector db + metadata store) in three weeks. It felt glorious — until production traffic hit and everything slowed down. Here’s what we learned the hard way: low-latency, high-recall retrieval at scale is not just about picking a vector DB. It’s an operational system with cost,…

  • Scaling LLM + Vector DB Systems: Lessons We Learned the Hard Way

    Introduction We shipped our first retrieval-augmented application (LLM + vector db + metadata store) in three weeks. It felt glorious — until production traffic hit and everything slowed down. Here’s what we learned the hard way: low-latency, high-recall retrieval at scale is not just about picking a vector DB. It’s an operational system with cost,…

  • Scaling LLM + Vector DB Systems: Lessons We Learned the Hard Way

    Introduction We shipped our first retrieval-augmented application (LLM + vector db + metadata store) in three weeks. It felt glorious — until production traffic hit and everything slowed down. Here’s what we learned the hard way: low-latency, high-recall retrieval at scale is not just about picking a vector DB. It’s an operational system with cost,…