Atik ptimizeroracle

Atik

I built data platforms, now LLM platforms: gateways, inference routing, serving on Kubernetes, agents on top.

Ondine: batch LLM processing for tabular data. Checkpointing, budgets, 100+ providers.

vllm#65569, measured: a proposal routes Linear layers to cuBLAS in vLLM's batch-invariant mode, on the claim that cuBLAS is already invariant there. True on every shape I tried but one: the gpt-oss-20b router (32 x 2880) changes bits with batch size, from 29 rows in bf16 and 2 in fp16. Tested on an H200 across every batch size from 1 to 1024; script and raw data.
LocalAI#12992, fixed: federated load balancing recorded a request only after the connection was open, so a burst of requests all saw the same idle node and went to it. Reported with a reproduction; fixed upstream.
llm-d-router#3379, merged: a pod that had never served a request got no traffic, so it never reported, so it stayed idle. Reproduced on a bench, then confirmed by the reporter on 4 x H100: idle pod in 10 of 12 runs before, 0 of 12 after.
aibrix#3297, fixed: least-latency routing scored a pod by its lifetime mean, so a pod that slowed down kept winning for a time that grows with its uptime. I measured the lag and derived when the choice flips; the fix now scores the last minute.
llm-d-router#3814, merged: the chart cut release and namespace to 40 characters each, so two namespaces with the same prefix got one cluster-wide RBAC name, and long names rendered invalid YAML. It was why a llm-d nightly failed. Now unique, with a regression test in the chart checks.

Data: where I come from. Spark, Kafka, Databricks, dbt, Polars, and the Hadoop era before that.
Inference: vLLM, SGLang, LiteLLM, Kubernetes. Queueing theory when the numbers stop making sense.
Agents: LangGraph, MCP, evals and tracing. Also Vauban, a reference LLM platform and agent lab I am building as a course: 65 controls, every claim tagged measured or gap. Private for now.

Python, SQL, Go.