Software Platform Architect

Saaki-Argus-and-Averil-Consulting; Bangalore Rural; Remote; INR 4,500,000 - 7,500,000; Full time

Job summary

Saaki, Argus and Averil Consulting seeks a Software Platform Architect to lead end-to-end software platform architecture for next-generation AI accelerators. You will bridge silicon, SDKs, compilers, runtimes, ML frameworks, and production inference, driving hardware-software co-design and platform performance.

You will own runtime, memory management, drivers, and framework integration, define roadmaps, and collaborate with silicon teams to optimize memory, scheduling, and PCIe bring-up.

Qualifications

• B.Tech/M.Tech or MS/Ph.D. in CS, EE, or related field.
• 15–20 years of experience in software architecture for HPC, GPU, NPU, or many-core AI accelerators.
• Ownership of a production software stack from silicon bring-up to customer deployment.
• Strong C/C++ systems programming and Python scripting.

Responsibilities

• Own the software platform architecture, including runtime scheduler, memory manager, drivers, APIs, compiler backend, and framework integration.
• Define software architecture roadmaps and SDK governance with performance benchmarks.
• Collaborate with silicon teams to optimize memory usage and accelerator performance.
• Lead software bring-up, PCIe validation, BF16 verification, and first LLM inference demonstration.

Skills

C/C++

Python

CUDA

HIP

OpenCL

MPI

OpenMP

Software architecture

HPC

Education

B.Tech/M.Tech or MS/PhD

Tools

LLVM

MLIR

CUDA toolkit

OpenCL SDK

PyTorch

DeepSpeed

Job description

Job Title: Software Platform Architect HardwareSoftware Co-DesignExperience: 1520 YearsLocation: Anywhere IndiaRole OverviewLead the end-to-end software platform architecture for next-generation AI accelerators, bridging silicon architecture, SDK development, compiler/runtime systems, ML frameworks, and production LLM inference. Drive hardwaresoftware co-design, platform performance, and software ecosystem development.Key ResponsibilitiesOwn the software platform architecture, including runtime scheduler, memory manager, drivers, low-level APIs, compiler backend, and framework integration.Define software architecture roadmaps, chip-generation compatibility, SDK governance, and platform performance benchmarks.Collaborate with silicon teams to optimise memory usage, MIMD scheduling, HBM access, and overall accelerator performance.Lead software bring-up planning and execution, from initial boot and PCIe validation to compute-unit testing, BF16 verification, and first LLM inference demonstration.Drive integration with PyTorch, vLLM, HIP, Transformers, and DeepSpeed.Establish porting strategies for CUDA, HIP, OpenCL, and MPI applications, including documentation and benchmarking.Lead SDK CI/CD automation, regression testing, release management, coding standards, and technical documentation.Mentor 510 engineers and collaborate with firmware, security, and external technology partners.Required QualificationsB.Tech/M.Tech or M.S./Ph.D. in Computer Science, Electrical Engineering, or a related field.1520 years of experience in software architecture for HPC, GPU, NPU, or many-core AI accelerator platforms.Proven ownership of a production software stack from silicon bring-up to customer deployment, covering compiler backend, runtime, driver interfaces, and ML framework integration.Strong C/C++ systems programming and Python scripting/tooling skills.Hands-on experience with parallel programming technologies such as CUDA, HIP, OpenCL, MPI, and OpenMP.Experience in hardwaresoftware co-design, heterogeneous workload porting, and performance optimisation.Preferred QualificationsExperience building software ecosystems for non-NVIDIA/AMD accelerator platforms.Familiarity with many-core architectures such as Tilera, Kalray, Graphcore IPU, custom HPC ASICs, or MIPS64 systems.Experience with LLVM compiler backends and/or MLIR.Ph.D. or advanced research experience in computer architecture, compilers, or HPC.