14 hours ago
Sunnyvale, CA, USASenior
Base Salary
$150k - $183k/yr
Responsibilities
- Build and train guardrail classifiers and token-level models for prompt injection, jailbreaks, unsafe content, sensitive data exposure, and policy violations.
- Design and tune tiered detection cascades to balance accuracy, false positives, throughput, and fixed per-request latency.
- Fine-tune encoder classifiers and taggers, adapt small decoder models, and use distillation to create deployable models.
- Quantize, compile, deploy, and tune models for GPU appliances using production inference servers, batching, concurrent execution, KV-cache configuration, and multi-stage pipelines.
- Research evasion techniques including obfuscation, encoding bypasses, dilution attacks, indirect injection, and multi-turn attacks, converting bypasses into training data and regression tests.
- Build realistic evaluation benchmarks, monitor deployed models for drift, and maintain multilingual personal-data, regulated-data, and natural-language policy detectors.
Requirements
- Strong Python and production PyTorch experience; experience with Go, Rust, C, or C++ for performance-critical paths is valuable.
- Demonstrated experience training, fine-tuning, and evaluating transformer models, including encoder classifiers, decoder language models, or both, using Hugging Face Transformers or an equivalent tool.
- Production experience with modern inference-serving systems such as Triton, vLLM, TensorRT-LLM, or TGI, including batching and memory tuning.
- Practical experience with quantization, distillation, pruning, or graph compilation while preserving model accuracy and reducing latency or memory use.
- Ability to design deployment-realistic test sets and reason about precision and recall when false positives affect legitimate traffic.
- Working knowledge of tokenization, text normalization, and Unicode handling as security-relevant attack surfaces.
- Familiarity with Docker, Kubernetes, experiment tracking, model versioning, reproducible training pipelines, and standard MLOps practices.
- Ability to deliver on schedule in an Agile environment and communicate across technical and non-technical teams.
- Preferred experience includes security or abuse detection modeling, LLM threat analysis, OWASP Top 10 for LLM Applications, LightGBM, XGBoost, NER, PII detection, multilingual classification, CUDA, GPU profiling, synthetic data, active learning, human-in-the-loop labeling, publications, open-source work, or adversarial ML and LLM security CTF/red-team experience.
Benefits
- Medical, dental, vision, life, and disability insurance.
- 401(k), 11 paid holidays, vacation time, sick time, and a comprehensive leave program.
- Eligibility to participate in the Fortinet equity program and potential bonus eligibility subject to company discretion.
- Full-time position with a stated U.S. base salary range of $150,000-$183,000.
- Must be authorized to work in the U.S. without sponsorship.
Categories
About Fortinet
Fortinet is a public cybersecurity company that builds network security hardware and software for enterprises, service providers, and governments. Its FortiGate firewalls, secure networking (SD-WAN), endpoint and cloud security, and threat intelligence subscriptions are sold via direct and channel partners. Founded in 2000 and headquartered in Sunnyvale, California, Fortinet trades on NASDAQ (FTNT) and serves more than 660,000 customers worldwide.