Goldman Sachs' Enterprise Technology Operations (ETO) team is looking to reduce the risk and cost of operating large-scale production environments by leveraging Machine Learning and Generative AI. The Production Runtime Experience (PRX) team specifically aims to streamline monitoring, alerting, automation, and workflows through advanced ML and GenAI applications.
Requirements
- 5+ years of software development in one or more languages (Python, C/C++, Go, Java); strong hands-on experience building and maintaining large-scale Python applications preferred.
- 3+ years designing, architecting, testing, and launching production ML systems, including model deployment/serving, evaluation and monitoring, data processing pipelines, and model fine-tuning workflows.
- Practical experience with Large Language Models (LLMs): API integration, prompt engineering, finetuning/adaptation, and building applications using RAG and tool-using agents (vector retrieval, function calling, secure tool execution).
- Understanding of different LLMs, both commercial and open source, and their capabilities (e.g., OpenAI, Gemini, Llama, Qwen, Claude).
- Solid grasp of applied statistics, core ML concepts, algorithms, and data structures to deliver efficient and reliable solutions.
- Proficiency building and operating on cloud infrastructure (ideally AWS), including containerized services (ECS/EKS), serverless (Lambda), data services (S3, DynamoDB, Redshift), orchestration (Step Functions), model serving (SageMaker), and infra-as-code (Terraform/CloudFormation).
Responsibilities
- Build agentic AI systems: Design and implement tool-calling agents that combine retrieval, structured reasoning, and secure action execution (function calling, change orchestration, policy enforcement) following MCP protocol.
- Productionize LLMs: Build evaluation framework for open-source and foundational LLMs; implement retrieval pipelines, prompt synthesis, response validation, and self-correction loops tailored to production operations.
- Integrate with runtime ecosystems: Connect agents to observability, incident management, and deployment systems to enable automated diagnostics, runbook execution, remediation, and post-incident summarization with full traceability.
- Safety, reliability, and governance: Build validator models, adversarial prompts, and policy checks into the stack; enforce deterministic fallbacks, circuit breakers, and rollback strategies; instrument continuous evaluations for usefulness, correctness, and risk.
- Scale and performance: Optimize cost and latency via prompt engineering, context management, caching, model routing, and distillation; leverage batching, streaming, and parallel tool-calls to meet stringent SLOs under real-world load.
- Build a RAG pipeline: Curate domain-knowledge; build data-quality validation framework; establish feedback loops and milestone framework maintain knowledge freshness.
- Raise the bar: Drive design reviews, experiment rigor, and high-quality engineering practices; mentor peers on agent architectures, evaluation methodologies, and safe deployment patterns.
Other
- Collaborate directly with users: Partner with production engineers, and application teams to translate production pain points into agentic AI roadmaps; define objective functions linked to reliability, risk reduction, and cost; and deliver auditable, business-aligned outcomes.
- Strong analytical problem-solving, ownership, and urgency; ability to communicate complex ideas simply and collaborate effectively across global teams with a focus on measurable business impact.
- A Bachelor’s degree (Masters/ PhD preferred) in a computational field (Computer Science, Applied Mathematics, Engineering, or in a related quantitative discipline), with 5+ years of experience as an applied data scientist / machine learning engineer.
