PhD in Computer Science, Machine Learning, Electrical Engineering, or equivalent experience in applied AI research.
8+ years of hands-on experience in machine learning systems, with 3+ years focused on LLMs or agentic AI.
Deep expertise in PyTorch and the Hugging Face ecosystem, including distributed training frameworks like DeepSpeed.
Strong foundation in supervised fine-tuning and reasoning-enhancement techniques.
Proven ability to design rigorous evaluation frameworks and automated benchmarks.
Experience in production ML monitoring and drift detection using statistical methods.
Track record of translating research into production systems, evidenced by publications or patents.
Responsibilities
Lead adaptation of foundation models for specialized engineering domains using advanced techniques.
Research and implement post-training methods for model optimization and reasoning enhancement.
Architect systems for multi-step planning and analytics over structured data sources.
Own evaluation strategy for LLM and agent systems, building automated frameworks and benchmarks.
Develop ML monitoring to detect feature and data drift, ensuring model reliability.
Build scalable pipelines for dataset construction and distributed training.
Set standards for experimental rigor and translate outcomes into publications and production designs.
Benefits
Comprehensive health and wellness benefits.
Financial benefits including potential bonuses and equity.
Support for professional development and growth opportunities.
Full Job Description
Descriptions & Requirements
Job Description and Requirements
What You'll Be Doing
Lead the adaptation of foundation models to specialized engineering domains like hardware description languages, RTL design, and EDA workflows, using supervised fine-tuning, LoRA/PEFT, and synthetic data pipelines that improve correctness and coverage
Research and implement advanced post-training methods including preference optimization, iterative alignment, and reasoning enhancement using Monte Carlo Tree Search, process reward models, and trajectory-based supervision
Architect agentic systems capable of multi-step planning, tool calling, self-correction, and analytics over structured data sources like CSV, JSON, Parquet, and SQL, with sandboxed execution and privacy-preserving designs
Own the evaluation strategy for LLM and agent systems, building automated frameworks with LLM-as-judge pipelines, domain-specific benchmarks, pass@k metrics, and regression suites that gate model releases
Develop production ML monitoring to detect feature drift, data drift, and performance degradation, using statistical tests like Kolmogorov-Smirnov, Wasserstein distance, and Jensen-Shannon divergence combined with feature importance
Build scalable pipelines for dataset construction, distributed multi-GPU training on NVIDIA A100/H100 clusters, checkpoint evaluation, and experiment tracking using PyTorch, Hugging Face, DeepSpeed, and FSDP
Set standards for experimental rigor across the team, including strong baselines, controlled ablations, reproducibility, error analysis, and honest reporting of negative results, and translate outcomes into publications, patents, and production-ready designs
The Impact You Will Have
Define the technical direction for domain-specific LLMs that generate verifiable hardware code, improving the productivity of engineering teams designing complex semiconductor systems
Enable reasoning-enhanced models that solve multi-step planning and coding tasks with measurably higher correctness, reducing iteration cycles for engineers working in EDA workflows
Deliver enterprise analytics agents that operate autonomously over heterogeneous structured data, automating analysis and reporting tasks that today require manual effort
Build evaluation frameworks that become the standard for how the team measures quality, gates releases, and maintains trust in production AI systems
Detect and remediate production ML drift before it impacts users, ensuring models remain reliable and accurate as data distributions shift over time
Grow the technical capability of the team through mentorship, design reviews, and by setting the example for what rigorous, reproducible AI research looks like in practice
Represent Synopsys AI research externally through publications and invention disclosures that advance the state of the art in domain-specific AI systems
What You'll Need
PhD in Computer Science, Machine Learning, Electrical Engineering, or equivalent practical experience in applied AI research and production ML systems
8+ years of hands-on experience building and shipping machine learning systems, with at least 3 years focused on LLMs, post-training, or agentic AI
Deep expertise in PyTorch, Hugging Face ecosystem (Transformers, PEFT, TRL, Datasets, Accelerate), and distributed training frameworks like DeepSpeed or FSDP
Strong foundation in supervised fine-tuning, parameter-efficient methods, preference optimization, and reasoning-enhancement techniques including search-based inference and reward modeling
Proven ability to design and implement rigorous evaluation frameworks, including automated benchmarks, LLM-as-judge pipelines, and statistical validation of model improvements
Experience building production ML monitoring, drift detection, and retraining pipelines using statistical methods and data-quality tools
Track record of translating research into production systems, with evidence through publications, patents, or shipped products that demonstrate both depth and engineering discipline
Who You Are
You do not trust a result until you have measured it three ways, ablated the components, and understood why it works, not just that it works
You can explain a complex tradeoff between model quality, inference cost, and latency to a VP in two sentences without losing the nuance, and you can defend your technical choices in a room full of skeptical engineers
You move between reading a new paper on process reward models in the morning and debugging a multi-GPU training pipeline in the afternoon without losing context or momentum
You push back when an evaluation is too weak, a baseline is missing, or a launch plan does not account for how the model will actually fail in production
You learn new technical domains quickly, whether that means understanding RTL simulation well enough to integrate it into a training loop or picking up a new statistical test to improve drift detection
You set the standard for the team through the quality of your code, the rigor of your experiments, and the clarity of your technical writing, and you grow the people around you by reviewing their work with the same care you apply to your own
The Team You'll Be Part Of
Your recruiter will share more about the team structure and mission during the interview process. Rewards and Benefits
We offer a comprehensive range of health, wellness, and financial benefits to cater to your needs. Our total rewards include both monetary and non-monetary offerings. Your recruiter will provide more details about the salary range and benefits during the hiring process.
In addition to the base salary, this role may be eligible for an annual bonus, equity, and other discretionary bonuses. Synopsys offers comprehensive health, wellness, and financial benefits as part of a competitive total rewards package. The actual compensation offered will be based on a number of job-related factors, including location, skills, experience, and education. Your recruiter can share more specific details on the total rewards package upon request. The base salary range for this role is across the U.S.
About Synopsys Inc
Synopsys, Inc. is an American electronic design automation company that focuses on silicon design and verification, silicon intellectual property and software security and quality. Products include logic synthesis, behavioral synthesis, place and route, static timing analysis, formal verification, hardware description language (SystemC, SystemVerilog/Verilog, VHDL) simulators, as well as transistor-level circuit simulation. The simulators include development and debugging environments which assist in the design of the logic for chips and computer systems.
