Senior Technical Architect (IN)

HCLTech

8–10 yrs India Full Time Work from office
HCLTech logo
Posted : today
Actively hiring

Job description

Lead the on-device AI execution stack for ARM-based SoCs. This pivotal role involves managing model conversion and quantization, runtime integration, and deploying advanced vision, LLM, and VLM workloads while adhering to strict edge power and memory constraints.

This is a senior-level position focused on optimizing artificial intelligence models for deployment on resource-constrained embedded systems. You will be instrumental in ensuring high performance and efficiency for AI applications running directly on devices.

Responsibilities

Spearhead the deployment and optimization of neural network models across NPU, GPU, and CPU backends utilizing embedded inference runtimes like LiteRT/TensorFlow Lite and ONNX Runtime.

Take ownership of the model conversion pipeline, including graph capture, operator mapping, quantization (PTQ and QAT), and validation.

Troubleshoot and resolve issues related to unsupported operators, graph partitioning, and fallback behaviors, collaborating with vendor toolchains to address accelerator limitations.

Enable on-device LLM and VLM workloads through weight quantization, KV-cache management, and optimization of prefill/decode processes to fine-tune memory footprint and token throughput.

Conduct comprehensive benchmarking and optimization of inference latency, throughput, memory bandwidth, and energy consumption, publishing reproducible performance data.

Integrate inference capabilities into real-time media and robotics pipelines, ensuring zero-copy tensor and buffer sharing for maximum efficiency.

Develop robust model deployment and regression tooling to meticulously track accuracy and performance across software releases.

Provide expert guidance to product and data-science teams on selecting model architectures that are optimal for edge accelerator capabilities.

Qualifications

A strong foundation in software engineering with 8-10 years of experience, including over 3 years specifically in on-device or edge AI deployment.

Proficiency in C++ and Python is essential.

Demonstrated hands-on experience with embedded inference runtimes such as LiteRT/TFLite, ONNX Runtime, ExecuTorch, TVM, or vendor NPU SDKs, including delegate or execution-provider integration.

Practical expertise in quantization techniques, including INT8/INT4, per-channel schemes, calibration, and accuracy recovery.

In-depth knowledge of model formats and conversion tooling, transitioning from frameworks like PyTorch/TensorFlow to deployable graph representations.

Experience in profiling performance on heterogeneous SoCs and understanding memory bandwidth as a primary performance bottleneck.

Solid understanding of CNN, transformer, and contemporary vision and language model architectures.

Familiarity with on-device LLM/VLM deployment concepts, such as llama.cpp class runtimes, speculative decoding, and paged KV cache is highly beneficial.

Experience with custom operator or kernel development for DSP/NPU/GPU using technologies like OpenCL, Vulkan compute, or DSP intrinsics is a plus.

Familiarity with compiler-level work, including MLIR, TVM, and graph-level optimization passes, is advantageous.

Experience with vision pipeline integration and implementing zero-copy paths from camera to inference is desirable.

Knowledge of MLOps for edge environments, including model versioning, A/B evaluation, and field accuracy monitoring, is a significant advantage.

Essential Skills

C++PythonEmbedded Inference RuntimesQuantizationModel ConversionSoC ProfilingCNNTransformer Architectures

Good to Have

LLM DeploymentCustom Operator DevelopmentMLIRTVMVision Pipeline IntegrationMLOps for Edge

Highlights

  • Actively hiring

More Details

RoleSenior Technical Architect (IN)
Employment TypeFull Time, Work from office

About the Company

HCLTech logo

HCLTech

IT Consulting

Senior Technical Architect (IN) at HCLTech | SkillMX | SkillMX