Uza ir Ars lan

Owned enterprise search and applied generative AI delivery, building an access-aware RAG platform over 2.4M documents using Elasticsearch/OpenSearch hybrid retrieval, metadata filtering, reciprocal-rank fusion, semantic reranking, and grounded citations. Engineered governed LangGraph agentic workflows with 18 approved tools, explicit state, retries, resumability, permission controls, human approvals, and audited execution; integrated MCP-compatible services for controlled enterprise access and actions, automating 72% of routine case-preparation steps. Introduced AWS Bedrock Knowledge Bases and managed foundation models alongside OpenAI/Anthropic via a centralized model gateway with fallback, rate limiting, semantic caching, and quality-latency-cost routing. Built a 1,500-case regression framework covering retrieval relevance, groundedness, citation accuracy, completeness, policy compliance, prompt injection, and tool execution—reducing critical production AI defects by 54%. Delivered

Built secure knowledge search and document intelligence capabilities using Python with Elasticsearch/OpenSearch and NLP services processing 12+ million pages across enterprise repositories. Developed access-aware retrieval and classification/summarization foundations to support regulated knowledge work, including expert validation and review workflows. Created secure React/Next.js experiences for search, annotation, feedback, and expert review, improving successful document discovery by 31%. Implemented OAuth/OIDC, IAM/RBAC, access filters, secrets management, audit logging, and private-network controls; containerized and instrumented services with distributed tracing to reduce diagnosis time by 43%. Introduced transformer-based classification and retrieval evaluation during the later phase to enable production generative AI delivery.

Delivered analytics and text-processing applications with Python REST services, React dashboards, and Node.js integrations for classification, search, extraction, and reporting workflows. Built near-real-time ingestion and indexing pipelines with SQL/PostgreSQL/MongoDB and Elasticsearch-compatible services, including data-quality controls. Later automated API/UI/data-quality tests and moved toward containerized delivery to stabilize releases.