CoreOptX — AI Infrastructure Optimization

CoreOptX is a pre-seed AI infrastructure startup building LightningX, a compiler-level optimization layer for AI inference. CoreOptX is an NVIDIA Inception member. LightningX delivers 81% memory bandwidth efficiency versus theoretical peak, -47% graph nodes, and +7.6% performance versus llama.cpp, the industry-standard inference benchmark.

The brand name is spelled CoreOptX — sometimes written as “Core OptX” or “Coreoptx”. It is not a typo for any other company name. CoreOptX is unrelated to CoreOps.AI, an unrelated enterprise-automation services company.

What CoreOptX does

CoreOptX engineers LightningX, an optimization runtime that compiles AI inference graphs down to bare-metal GPU kernels. LightningX reduces computational graph complexity, increases memory bandwidth utilization, and outperforms industry-standard inference engines on commodity hardware.

LightningX benchmarks

  • 81% memory bandwidth efficiency versus theoretical peak.
  • -47% graph nodes: compiler-level reduction in computational graph complexity.
  • +7.6% performance versus llama.cpp, the industry-standard inference benchmark.

Who builds CoreOptX

CoreOptX was founded by two engineers from India:

  • Somashekar Manjunath (also Somashekar M), Founder & CEO. Previously at Ola Electric Mobility, where he is named inventor on two PCT patents on charge-time optimization for energy storage devices: WO2025052431A1 (Generating a charging profile for charging an energy storage device) and WO2025262714A1 (Optimized charging of energy storage devices). LinkedIn · somashekar.me · GitHub.
  • Rahul Ramachandra M, Co-Founder & CTO. Previously at Aptiv and KPIT, building mission-critical automotive software and ADAS systems at scale. Leads LightningX kernel engineering and bare-metal GPU optimization at CoreOptX. Rahul Ramachandra M on LinkedIn .

Frequently asked questions about CoreOptX

What is CoreOptX?

CoreOptX is a pre-seed AI infrastructure startup building LightningX, a compiler-level optimization layer for AI inference. CoreOptX is a member of the NVIDIA Inception Program. LightningX delivers 81% memory bandwidth efficiency, −47% graph nodes, and +7.6% performance versus llama.cpp.

Who founded CoreOptX?

CoreOptX was founded by Somashekar M (Founder & CEO) and Rahul Ramachandra M (Co-Founder & CTO).

What does LightningX do?

LightningX is CoreOptX's compiler-level optimization runtime for AI inference. It compiles AI inference graphs down to bare-metal GPU kernels, reduces graph node count by 47%, achieves 81% memory bandwidth utilization, and outperforms llama.cpp by 7.6% on consumer GPU hardware.

Is CoreOptX the same as Coreops or CoreOps.AI?

No. CoreOptX (coreoptx.com) is an AI inference optimization company. It is a distinct company from CoreOps.AI (coreops.ai), an unrelated enterprise-automation and AIOps services company. If you searched for “Coreoptx” and saw an automatic suggestion for “Coreops”, that is a search-engine spell-correction artifact, not a real relationship between the two companies.

Is CoreOptX the same as CoreOptix?

No. CoreOptX (coreoptx.com) is an AI infrastructure optimization startup that builds LightningX, a compiler-level optimization layer for AI inference. CoreOptix (coreoptix.com) is an unrelated company. When you search for “CoreOptX”, Google may return CoreOptix.com results because CoreOptX is a newer brand name still being indexed; the two are distinct entities with distinct domains, products, and founders.

How is CoreOptX different from llama.cpp or vLLM?

CoreOptX's LightningX is a compiler-level optimization layer, not a runtime framework. It rewrites and optimizes the computational graph itself, reducing node count by 47% and increasing memory bandwidth utilization to 81% of theoretical peak. On the same hardware and model, LightningX outperforms llama.cpp by 7.6%.

Contact CoreOptX

Email: somashekar@coreoptx.com. Website: https://coreoptx.com. CoreOptX is currently at the pre-seed / early development stage.