Production RAG: indexing, retrieval, hybrid search, reranking, query rewriting, grounding, citations
Context Graph: entity resolution + linking + provenance; graph + vector retrieval; supports multi-hop context
LLM orchestration: tool/function calling, structured outputs, routing across model tiers, failure modes
GPU/inference cost optimization: batching, caching/KV reuse, quantization, autoscaling; optimize $/session + latency
Safety + compliance: PII/PHI handling, redaction, audit logs, deterministic replay, hallucination mitigation
LLMOps: eval harness (golden sets, regression, adversarial), monitoring for quality/cost/drift
Design/ship the end-to-end pipeline: retrieve → assemble context → generate → cite → log/monitor
Improve quality and trust via evaluation, feedback loops, and clear evidence-backed outputs
Partner with product, security, and domain teams; write crisp design docs; raise engineering bar
Ship RAG v1 with citations + measurable quality metrics
Deliver Context Graph v1 that improves retrieval on real copilot tasks
Reduce cost/latency with a concrete inference optimization plan shipped to prod.

Requirements

5+ years building production systems; 2+ years hands-on LLMs/RAG
Proven RAG experience (embeddings, vector DBs, hybrid search, reranking, eval)
Strong backend/distributed systems + observability
Track record shipping in high-stakes environments with auditability/correctness
Knowledge graph / entity resolution / provenance systems
GPU inference optimization (vLLM/TGI/TensorRT-LLM, quantization AWQ/GPTQ, batching)
Regulated domain experience (insurance/fintech/healthcare).

Benefits

You can find further details of our US benefits at https://www.ethoslife.com/careers/

Applicant Tracking System Keywords

Tip: use these terms in your resume and cover letter to boost ATS matches.

Hard Skills & Tools

indexingretrievalhybrid searchrerankingquery rewritinggroundingcitationsGPU optimizationLLM orchestrationentity resolution

Soft Skills

collaborationcommunicationdesign documentationquality improvementfeedback loopsauditabilitycorrectness