AI Systems Research Intern - Optimize, Deploy, Benchmark

Naïve; Mountain View (CA); On-site; USD 32,000 - 52,000; Full time

Job summary

Naïve is an autonomous company builder that deploys AI employees to run entire companies. We’re a $120M company backed by prominent investors.

We’re building cost-effective, scalable AI systems that can operate fleets of agents with lower costs and higher reliability. As an intern, you’ll research and ship the cost/performance frontier for thousands of AI agents, optimize local model inference, and develop routing strategies to choose the cheapest model that can do the job.

Qualifications

• Strong systems and ML engineering background with Python and PyTorch.
• Deep understanding of how transformers run: attention, KV cache, memory bandwidth.
• Built and shipped a real project—OSS, side project, or internship.
• Experience with LLMs as tooling and objects of study via API calls and prompts.
• Move fast, take feedback, push back when you're right.

Responsibilities

• Research and ship cost/performance frontier for AI agents.
• Optimize local/self-hosted model inference and related techniques.
• Build model routing to select cheapest capable model.
• Benchmark and deploy across hardware (GPUs, edge, on-prem, accelerators).
• Push on agent infrastructure: orchestration, caching, context management, parallelization.
• Prototype recursive self-improvement loops.
• Own a research question end-to-end, from framing to production.

Skills

High agency

Systems ML engineering

Python & PyTorch

Profiling & optimization

Ship without hand-holding

Education

CS/EE/Math student

Tools

vLLM

TensorRT-LLM

SGLang

llama.cpp

Job description

Naïve is an autonomous company builder that deploys AI employees to run entire companies. We’re a $120M company backed by prominent investors.

We’re building cost-effective, scalable AI systems that can operate fleets of agents with lower costs and higher reliability. As an intern, you’ll research and ship the cost/performance frontier for thousands of AI agents, optimize local model inference, and develop routing strategies to choose the cheapest model that can do the job.