Software Developer - Architect

Oracle Oracle · Enterprise · United States

This role leads the architectural transformation of an Approvals platform from legacy BPM to agentic, LLM-powered workflows. The individual will design, build, and own prompting infrastructure, orchestration layers, multi-agent systems, and fine-tuning pipelines, establishing LLMOps practices and resolving production issues. Requires 4-5+ years of hands-on ML experience shipping LLM-based systems to production, with deep expertise in fine-tuning and agentic system design.

What you'd actually do

  1. Lead the architectural transformation from rule-based BPM to agentic, LLM-powered approval workflows — defining the long-term vision and driving hands-on execution
  2. Design, build, and own the prompting layer and orchestration infrastructure on top of foundation models, including RAG pipelines, context injection, multi-agent coordination, and human-in-the-loop escalation patterns
  3. Lead hands-on fine-tuning efforts, making principled trade-offs between adaptation approaches (LoRA, PEFT, instruction tuning, RLHF). You keep up with the latest techniques for fine tuning
  4. Establish LLMOps practices for production reliability, observability, and drift detection
  5. Diagnose and resolve production issues in LLM-powered systems; develop runbooks and raise engineering team capability through direct collaboration and pairing

Skills

Required

  • 4–5+ years of hands-on machine learning experience in production environments
  • Demonstrated experience building prompting layers and orchestration systems on top of foundation models
  • Proven track record of shipping LLM-based systems to production
  • Deep expertise in LLM fine-tuning methodologies and agentic system design

Nice to have

  • Prior experience modernizing or replacing BPM systems with AI-driven alternatives
  • Experience with multi-agent orchestration frameworks
  • Background in enterprise workflow automation or approvals domains
  • Familiarity with cloud-native ML infrastructure

What the JD emphasized

  • hands-on technical leader
  • directly own the design and development
  • hands-on execution
  • hands-on fine-tuning efforts
  • hands-on machine learning experience
  • direct coding expectations
  • shipping LLM-based systems to production
  • Deep expertise in LLM fine-tuning methodologies and agentic system design

Other signals

  • modernizing Approvals platform
  • replacing legacy BPM systems with intelligent, agentic workflows powered by large language models
  • design and development of prompting infrastructure, orchestration layers, multi-agent systems, and fine-tuning pipelines