About Nihilus Development

What the AIs Thought It Took

We showed the Nihilus codebase to four major AI systems and asked them to estimate team size, timeline, funding, and corporate backing based on the technical sophistication, production infrastructure, and benchmark results.

Here's what they concluded:

GEMINI 3 PRO
TEAM SIZE: 4-8 core engineers + 5-10 extended contributors (10-15 total)
TIMELINE: 2.0-3.5 years
FUNDING: $20 Million (Seed/Series A)
BACKING: NVIDIA (or NVIDIA-backed)
"The technical complexity points toward a small, highly specialized, elite team... This type of deep low-level optimization is typically achieved by core experts. The codebase is clearly C++ and CUDA-focused, emphasizing maximum performance, compile-time techniques, and memory layout optimization."
"If this project is real, it will fundamentally change the landscape of LLM inference serving."
GROK 4
TEAM SIZE: 5-15 engineers
TIMELINE: 1.5-3 years
FUNDING: $5-50 Million (seed to Series A)
BACKING: Likely NVIDIA
"This level of innovation (custom CUDA kernels beating CUTLASS in irregular shapes, full transformer rewrite with novel fusions, hybrid static/dynamic dims) suggests a small but elite group of systems/C++/CUDA experts."
"The maturity—single-TU design, extreme code hygiene (-Weverything -Werror), production-ready features—indicates sustained full-time effort, not a side project."
DEEPSEEK V3
TEAM SIZE: 5-12 core engineers/researchers
TIMELINE: 12-24 months of intensive development
FUNDING: $2-5 Million (seed/Series A)
BACKING: Most likely NVIDIA
"2-3 CUDA/low-level optimization experts, 2-3 C++/compiler experts, 1-2 systems engineers, 1-2 researchers, 1 technical lead/architect overseeing the cohesive design."
"This is production-grade. Not a research prototype. The memory bandwidth reductions (99%+) are extraordinary if accurate."
CLAUDE SONNET 4.5
TEAM SIZE: 8-15 people (revised upward)
TIMELINE: 24-36 months
FUNDING: $8-15 Million (Series A or well-funded seed)
BACKING: NVIDIA (95% confidence)
"This isn't just research—this is battle-tested production infrastructure. Alert thresholds for 1000-2000 concurrent connections indicate this is designed for serious production load."
"The 99.98% memory bandwidth reductions combined with production-grade monitoring infrastructure means: This is ready for deployment NOW. Someone is using it or about to deploy it."
THE ACTUAL ANSWER
TEAM SIZE: 1 person + Claude Sonnet 4.5
TIMELINE: 6 months (Jun 2025 - Dec 2025)
FUNDING: $0 external funding
BACKING: None (built independently)
WORKING ARRANGEMENT: Built while employed elsewhere

What every AI estimated would require 5-15 elite engineers,
years of development, and millions in funding—

Was built by one human and one AI
working at matched bandwidth
for 40+ hours per week.

Even the AIs can't believe what's possible
with AI-human collaboration at this intensity.