Job Description

Sfiniti AI is building private/local AI infrastructure and native structured-
compute primitives for efficient inference, robotics, and edge systems. The
current stack includes measured native GPU, CPU, ROCm, and MLX paths, exact
fallback behavior, runtime integration with private infrastructure, and
reviewer-grade evidence packages. We are now moving from research evidence
into production hardening.

This is not a ticket-taking role. The work is discovery-driven systems
engineering: audit what exists, preserve what works, identify false positives,
harden native paths, improve runtime boundaries, and help turn measured wins
into a product.

Responsibilities:

* Review and harden native C++/Python/GPU runtime code
* Improve benchmark methodology, profiling, and reproducibility
* Help package private compute primitives into clean SDK/runtime surfaces
* Work on inference runtime integration, local AI serving, and distributed execution
* Preserve exact fallback paths and fail-closed behavior
* Build tests, gates, documentation, and production-readiness evidence
* Help decide what should stay private, what can be public, and what is ready for customers

Strong candidates may have experience with:

* CUDA, ROCm/HIP, MLX/Metal, or low-level GPU performance
* PyTorch extensions, cuBLAS/BLAS, Triton, llama.cpp, vLLM, TensorRT-LLM, or KV-cache systems
* C++, Python, Linux, profiling, benchmarking, and distributed systems
* Hardware/software co-design, FPGA/ASIC, robotics, or edge AI
* Research-to-production engineering where requirements are incomplete and evidence matters

You do not need to know every area. You do need strong systems judgment,
comfort with ambiguity, and the ability to turn messy research into reliable
engineering without rewriting everything from scratch.

For exceptional candidates, we are also open to starting with a paid technical
review, advisory engagement, or part-time trial before converting to full-time
if there is mutual fit.