Machine Learning Engineer

Title: Machine Learning Engineer – LLM Fine-Tuning (Verilog/RTL Applications) Location: Hybrid, San Jose, CA (3 days onsite) locals Visa: GC/USC Duration: 18+MONTHS MOI: Video Job Description: JOB DETAILS Our client is developing privacy-preserving LLM capabilities that enable hardware design teams to reason over Verilog/System Verilog and RTL artifacts — including code generation, refactoring, lint explanation, constraint translation, and spec-to-RTL assistance. They are seeking a Staff-level Machine Learning Engineer to lead a small, high-impact team responsible for fine-tuning and productizing LLMs for these workflows in a strict enterprise data-privacy environment. You don’t need to be a Verilog/RTL expert to start; curiosity, drive, and deep LLM craftsmanship matter most. Any HDL/EDA fluency is a strong plus. Key Responsibilities: Own the technical roadmap for Verilog/RTL-focused LLM capabilities — from model selection and adaptation to evaluation, deployment, and continuous improvement. Lead and mentor an applied science/engineering team; set direction, review code/designs, and raise the bar on experimentation speed and reliability. Fine-tune and customize models using LoRA/QLoRA, PEFT, instruction tuning, and RLAIF with HDL-specific evaluation metrics (compile/simulate pass rates, constrained decoding, “does-it-synthesize” checks). Design and secure ML pipelines on AWS: leverage Bedrock (Anthropic and other FMs), SageMaker, or EKS for training/inference with strong privacy boundaries (S3 + KMS, PrivateLink, IAM least-privilege, CloudTrail, Secrets Manager). Deploy low-latency inference environments (vLLM/TensorRT-LLM) with autoscaling, blue-green rollouts, and canary testing. Build automated regression and evaluation suites for HDL compilation/simulation with MLflow or Weights & Biases tracking. Collaborate with Hardware Design, CAD/EDA, Security, and Legal to prepare compliant datasets and define acceptance gates. Drive integration of LLMs into internal developer tools, retrieval systems (RAG), and CI/CD pipelines. Foster a secure-by-default culture and mentor ICs on best practices for fine-tuning, reproducibility, and model governance. Minimum Qualifications: 10+ years of total engineering experience, including 5+ years in ML/AI or distributed systems, and 3+ years working directly with transformers/LLMs. Proven record shipping production LLM-powered features and leading at the Staff level. Hands-on expertise with PyTorch, Hugging Face Transformers/PEFT/TRL, DeepSpeed/FSDP, and constrained decoding. Deep AWS experience with Bedrock (Anthropic, Guardrails, Knowledge Bases, Runtime APIs) and SageMaker (Training, Inference, Pipelines), plus S3, EKS, IAM, KMS, CloudTrail, PrivateLink, and Secrets Manager. Strong fundamentals in testing, CI/CD, observability, and performance tuning; Python required (Go/Java/C++ a plus). Excellent cross-functional leadership and technical communication skills. Preferred Qualifications: Familiarity with Verilog/SystemVerilog, RTL workflows (lint, synthesis, timing closure, simulation, formal verification). Experience with grammar-constrained decoding or AST-aware tokenization for code models. RAG at scale over code/specs; function-calling for code transformation. Inference optimization (TensorRT-LLM, KV-cache tuning, speculative decoding). Experience with SOC2/ISO/NIST frameworks, red-teaming, and secure evaluation data handling. Data anonymization, DLP scanning, and IP-safe code de-identification. Success Metrics 90 Days: Establish secure AWS training/inference environments and an HDL-aware evaluation harness. Deliver an initial fine-tuned model with measurable HDL improvements. 180 Days: Expand fine-tuning coverage (Bedrock/SageMaker), add retrieval and constrained decoding, and deploy inference with reliability SLOs. 12 Months: Demonstrate measurable productivity gains for RTL teams (defect reduction, lint improvements, faster review cycles). Build a stable, compliant MLOps foundation for continuous LLM improvement. Security & Privacy by Design: No public internet calls; workloads isolated in private AWS VPCs. All data encrypted with KMS; access tightly controlled via IAM and CloudTrail auditing. Pipelines enforce minimization, DLP scanning, and reproducibility with model cards and data lineage.