23 hours ago
Shenzhen, ChinaStaff+
Responsibilities
- Design, develop, and productionize AI safety detectors for prompt injection, jailbreaks, personally identifiable information, sensitive data, data leakage, harmful content, policy violations, system prompt leakage, and agent or tool-use risks.
- Build detector pipelines using rule-based detection, pattern matching, NLP/ML classifiers, embedding-based models, LLM-as-a-Judge evaluators, and hybrid strategies with confidence scoring and policy-driven actions.
- Develop and optimize guardrail model evaluation workflows, including test-set creation, data labeling, fine-tuning, model validation, threshold calibration, precision/recall optimization, regression testing, and continuous improvement.
- Evaluate, benchmark, integrate, and productionize industry, open-source, and third-party AI safety solutions, including build-versus-buy assessments.
- Design end-to-end AI safety protection patterns for RAG systems, coding assistants, conversational AI, and agent applications across input, retrieval, generation, output, logging, and tool-execution layers.
- Build AI safety observability and governance capabilities, including telemetry, tracing, dashboards, evaluation metrics, audit trails, feedback loops, and operational controls.
- Partner with AI platform, infrastructure, security, and application teams on performance, scalability, detector lifecycle management, and reliable production rollouts.
Requirements
- Experience developing first-party AI safety detectors for prompt injection, jailbreaks, personally identifiable information, data leakage, toxicity, policy violations, or agent safety.
- Experience building fine-tuning pipelines for safety-oriented large or small language models, including dataset curation, labeling strategy, training workflow implementation, evaluation design, and production validation.
- Experience evaluating, integrating, and productionizing third-party or open-source detector solutions based on quality, latency, cost, operational fit, and governance requirements.
- Experience making build-versus-buy decisions and taking internally developed or externally sourced detector solutions through production rollout and ongoing operation.
- Experience with AI evaluation or observability tools such as Langfuse, LangSmith, Arize, Phoenix, or similar platforms.
- Familiarity with data loss prevention, secret scanning, document classification, or enterprise information protection tooling.
- Experience with adversarial testing, red teaming, or continuous attack simulation for AI systems.
Benefits
- Flexible working arrangements and opportunities for continuous professional development and career growth.
- Inclusive and diverse workplace with equal-opportunity employment practices.
Tech Stack
Phoenix
Categories
About HSBC
HSBC is a global bank that provides retail, commercial and investment banking, payments, and wealth management to individuals, SMEs, and multinationals. It earns fees, interest income, and trading revenues through branches and digital platforms across major markets. Headquartered in London, it serves over 40 million customers in 58 countries and is listed in London, Hong Kong, New York, and Bermuda.
