Joshua Owens

Architected a secure RAG web platform using Python, FastAPI, Next.js, REST APIs, embeddings, hybrid vector search, and document-level RBAC to reduce knowledge-retrieval time by 42%. Engineered a provider-independent LLM orchestration service integrating OpenAI, Anthropic, and Azure OpenAI with retries, token controls, and fallback rules to reduce integration effort by 35%. Established automated evaluation and CI/CD quality gates for groundedness, citation accuracy, safety, latency, and cost, reducing validation from three days to under six hours. Led architecture reviews and delivery planning across engineering, AI, cloud, security, and product stakeholders, improving retrieval precision by 18% and delivering three releases on schedule.

Orchestrated a distributed multimodal task-assignment backend using Python, microservices, transactional workflows, and REST APIs, increasing assignment throughput by 32% and reducing queue wait time by 27%. Hardened workflow services with idempotency, retries, backpressure, and event-driven state management, decreasing stalled or duplicate tasks by 38% during high-volume processing. Developed an LLM evaluation platform for pairwise ranking, rubric-based reviews, dataset exports, and model benchmarking, shortening evaluation dataset release cycles by 35%. Partnered with research, product, operations, and quality teams across four launches to translate requirements into scalable production workflows and reduce late-stage changes by 20%.

Modernized recurring billing and failed-payment recovery using Python, Java, Node.js, event-driven microservices, and configurable retries, improving eligible-payment recovery by 5.4%. Safeguarded financial state transitions with idempotency keys, durable event processing, invariant validation, and controlled retries, reducing duplicate workflow execution by more than 90%. Automated transaction and settlement reconciliation with Python normalization, SQL, layered matching, and checkpointed batch processing, resolving 93% of standard records and cutting processing time by 48%. Instrumented payment services with distributed tracing, dashboards, and discrepancy alerts while collaborating with product, risk, compliance, and operations teams, reducing mean time to resolution by 35%.