SoFi is a next-generation financial services company and national bank using innovative technology to help members reach their financial goals. The Senior Staff AI Engineer role is critical in setting the technical direction and delivering complex AI initiatives within the risk organization, focusing on developing best-in-class AI-based solutions for risk management and compliance.
Responsibilities:
- Define the long-term technical architecture and strategy for our next-generation AI platform, particularly focusing on robust, scalable agentic frameworks and LLM deployment patterns
- Architect and standardize the use of graph-based LLM orchestration, leveraging expert-level mastery of LangGraph to solve highly complex, multi-stage reasoning problems at scale
- Develop robust, persistent infrastructure for agentic state management, ensuring that long-running agent workflows maintain context and reliability across distributed nodes and regional failovers
- Pioneer and institutionalize advanced parameter-efficient fine-tuning (PEFT) and compression techniques to maximize model performance and minimize operational costs across the organization
- Support the development of a unified model serving platform designed to host internally fine-tuned and custom-trained models to ensure high-throughput, low-latency inference across diverse hardware footprints
- Define and enforce high standards for AI operationalization, requiring mastery in designing and deploying comprehensive AI observability solutions and advanced tracing/testing frameworks that guarantee production quality, compliance, and reliability
- Mentor senior and junior AI Engineers, elevating the overall engineering quality
- Coordinate with cross-functional teams to distill specific requirements, project roadmaps, and ensure accurate and on-time project deliveries
- Stay up-to-date with the latest trends and advancements in GenAI, LLMs, and NLP, evaluating and experimenting with new techniques and tools to push the boundaries of AI innovation in the banking sector