• Design, build, and operate agentic AI systems end-to-end—from concept to production. • Work on multi-agent orchestration, Retrieval-Augmented Generation (RAG), evaluation frameworks, and AI guardrails to build safe, reliable, and high-performing systems. • Collaborate cross-functionally with product, ML, and design teams—bringing ideas to life through strong engineering execution, clear communication, and a low-ego, problem-solving mindset. • Design and implement Retrieval-Augmented Generation pipelines to ground LLMs in enterprise or domain-specific data. • Make strategic decisions on chunking strategy, embedding models, and retrieval mechanisms to balance context precision, recall, and latency. • Work with vector databases (Qdrant, Weaviate, pgvector, Pinecone) and embedding frameworks (OpenAI, Hugging Face, Instructor, etc.). • Diagnose and iterate on challenges like chunk size trade-offs, retrieval quality, context window limits, and grounding accuracy—using structured evaluation and metrics. • Establish comprehensive evaluation frameworks for LLM applications, combining quantitative (BLEU, ROUGE, response time) and qualitative methods (human evaluation, LLM-as-a-judge, relevance, coherence, user satisfaction). • Implement continuous monitoring and automated regression testing using tools like LangSmith, LangFuse, Arize, or custom evaluation harnesses. • Identify and prevent quality degradation, hallucinations, or factual inconsistencies before production release. • Collaborate with design and product to define success metrics and user feedback loops for ongoing improvement. • Implement multi-layered guardrails across input validation, output filtering, prompt engineering, re-ranking, and abstention (“I don’t know”) strategies. • Use frameworks such as Guardrails AI, NeMo Guardrails, or Llama Guard to ensure compliance, safety, and brand integrity. • Build policy-driven safety systems for handling sensitive data, user content, and edge cases with clear escalation paths. • Design and operate multi-agent workflows using orchestration frameworks such as LangGraph, AutoGen, CrewAI, or Haystack. • Coordinate routing logic, task delegation, and parallel vs. sequential agent execution to handle complex reasoning or multi-step tasks. • Build observability and debugging tools for tracking agent interactions, performance, and cost optimization. • Evaluate trade-offs around latency, reliability, and scalability in production-grade multi-agent environments.