AI Engineer

General Summary

Realtek's Ameba AIoT product line delivers Wi-Fi/BT MCU and edge-AI SoCs used in smart home, security, and connected consumer devices worldwide.

We are looking for an AI Engineer to advance AI capabilities across the Ameba software stack — from AI-assisted development tooling that lets developers build on Ameba through natural language, to vision and speech applications that run on our NPU-enabled SoCs.

You will extend our Model Context Protocol (MCP) toolchain so that AI agents can build, deploy, and validate directly on Ameba hardware; bring vision and speech models onto our edge-AI platforms; and turn the results into reference designs that customers and ecosystem partners can adopt.

This is a role for someone who works at the intersection of applied AI and real hardware — comfortable moving a model from training framework, through quantization, onto a memory-constrained device, and then proving with data that it works. You will collaborate closely with our firmware and platform software teams, who own the RTOS, drivers, and low-level platform integration.

Key Responsibilities

• AI Development Toolchain. Extend the Ameba SDK Development MCP server with new agent-facing capabilities — model deployment, inference control, camera capture, and NPU performance reporting — so that AI coding agents can drive the full development loop on Ameba hardware.

• Benchmarking and Quality Measurement. Design and run end-to-end evaluation of the AI-assisted development workflow: task success rate, iteration count, and failure-mode analysis across multiple Ameba SoCs and AI clients. Use the results to guide engineering priorities and improve developer experience.

• Edge AI Model Enablement. Port, quantize, and optimize vision and speech models for Ameba NPU platforms. Produce and maintain a benchmarked model zoo covering inference latency, memory footprint, power consumption, and post-quantization accuracy.

• Bring-Your-Own-Model Workflow. Build and document the end-to-end path for customers to bring their own trained models onto Ameba — conversion, INT8 quantization, accuracy recovery, and on-device validation — including troubleshooting guides for common failure modes.

• Vision Reference Designs. Develop application-level reference designs on Ameba NPU SoCs for scenarios such as human and object detection and commercial video surveillance, validated against real-world performance targets.

• Speech and Voice Applications. Prototype hybrid on-device / cloud voice architectures — on-device wake word and audio front-end, cloud ASR, LLM, and TTS — and characterize the end-to-end latency budget that determines user experience. Benchmark speech technology components against measured accuracy, memory, and power targets.

• Partner and Customer Enablement. Serve as the technical counterpart for international AI/ISV ecosystem partners and support customer teams in adopting Ameba AI capabilities, including reference code, documentation, and issue resolution.

Minimum Qualification

• Master's degree in Electrical Engineering, Electronic Engineering, Computer Science, or a related field, with 5+ years of relevant AI/ML engineering experience.

• Demonstrated end-to-end ownership of machine learning systems in production — from model training and fine-tuning, through optimization and quantization, to deployed inference serving live business operations with measurable outcomes. Research-only or proof-of-concept-only experience does not meet this requirement.

• Experience building and maintaining the pipelines, APIs, or internal platforms that other engineers or end users depend on — not solely notebook-based model development.

• Practical experience with model optimization and deployment toolchains such as ONNX, OpenVINO, or TensorRT, including quantization of models for inference on constrained hardware and systematic recovery of post-quantization accuracy.

• Strong proficiency in Python for model development, tooling, and service implementation.

• Hands-on experience with deep learning frameworks (PyTorch, TensorFlow) for training and fine-tuning models.

• Experience leading technical delivery, either owning a project end to end or directing a small team through a full delivery cycle.

• Ability to read and interpret C source code well enough to understand SDK examples, trace data flow, and communicate precisely with firmware engineers. (Writing production firmware is not required for this role.)

• Professional working proficiency in English, both written and verbal.

Preferred Qualifications

• Experience building developer-facing tools or SDKs consumed by external developers.

• Experience deploying models to embedded, MCU-class, or NPU-accelerated hardware.

• Experience with computer vision model development and deployment (OpenCV, YOLO-family object detection, anomaly detection).

• Experience with speech and audio AI: ASR model fine-tuning, text-to-speech synthesis, or conversational systems.

• Experience building LLM-based applications: retrieval-augmented generation (RAG), agent frameworks, and systematic evaluation of AI system quality.

• Familiarity with the Model Context Protocol (MCP) or comparable AI agent tooling standards.

• Experience delivering technical enablement to external customers or partners.

• Track record of open-source contribution, technical publication, or competitive AI/ML achievements.