Vijay Karthik
Distributed systems enthusiast!
Engineering
How Going Back to the Stone Age Can Cut AI Costs
I taught my AI sub-agents to talk like cavemen. Alert investigations got 23% cheaper, 20% faster with no impact on quality
Vijay Karthik
AI
How We Made It Possible to Analyze Every AI Agent Span
How we made 7-day AI agent trace queries 30x faster, from 60s to under 2s, by changing how Oodle stores large prompts: out-of-band dictionaries, lazy loading.
Vijay Karthik
AI
You Can't Sample Your Way to Reliable Agents
Sampling 1% of agent traces for LLM-as-a-judge evals misses rare failures. How Oodle checks every trace with sentiment, anomaly detection, and code evaluators.
Vijay Karthik
Engineering
How Oodle Keeps Observability Fast at Scale
How Oodle keeps metrics queries fast at scale: durable S3 storage, AWS Lambda fan-out for bursty queries, compaction, and a columnar PromQL engine.
Vijay Karthik
News
Introducing Oodle AI: The fastest, most cost-efficient observability platform
Oodle AI is a serverless, fully managed observability platform, a drop-in Prometheus/ELK replacement at ~1/10th the cost.
Kiran Gollu, Vijay Karthik
Engineering
Go faster! Golang performance optimization in production
Golang performance optimization from production: buffer reuse, typed pools, GOMEMLIMIT, PGO. One slice fix cut 9.8 GB of allocations to 1 MB.
Vijay Karthik
Engineering
Building a High-Performance, Low-Cost Metrics Observability System
How Oodle built a serverless, Prometheus-compatible metrics system on S3 and AWS Lambda: storage split from compute, 10x cheaper storage, no AZ egress costs.
Vijay Karthik