Renz Patdu

Worked on enterprise AI search and knowledge-assistant capabilities for Jira and Confluence using React, TypeScript, Java/Spring Boot, Python services, OpenSearch, Redis, Docker, and Kubernetes. Implemented permission-aware hybrid retrieval (BM25 + vectors), metadata filters, reranking, source citations, and authorization checks to improve offline top-result relevance while preserving tenant isolation. Optimized LLM orchestration and conversational APIs with caching, batching, async processing, and query tuning to reduce p95 latency while keeping traceability and audit controls. Collaborated with product, security, platform, and machine learning teams on evaluation and production retrospectives to reduce post-release defects.

Led full-stack legal-medical document intelligence workflows with React, TypeScript, Python/FastAPI, PostgreSQL, Redis, Celery, Azure services, OCR, RAG, and LLM integrations. Built secure case and document-management workflows including upload/preview, processing status, RBAC, encrypted storage, audit trails, background jobs, and human review to reduce case setup and organization time. Developed document-grounded demand generation using Azure OpenAI, embeddings, approved templates, prompt versioning, output validation, and attorney approval to significantly reduce drafting time. Engineered AI case-summary and medical-chronology processing with Azure AI Document Intelligence, pgvector, extraction rules, confidence thresholds, and reviewer flags, and integrated external case systems via OAuth2, REST APIs, webhooks, Azure Service Bus, idempotency, retries, and reconciliation.

Progressed from healthcare software development into applied machine learning and clinical NLP for claims review, anomaly detection, medical coding, and care-management products using Java, Python, React, PostgreSQL, Redis, and cloud-native services. Built claims-review APIs and analyst dashboards with audit controls and integrations to reduce review time and automate initial validation. Developed ML pipelines and anomaly models (XGBoost, random forests, logistic regression, isolation forests) with Spark and Airflow to improve inaccurate-claim detection at scale. Implemented explainable inference workflows for reviewer feedback and clinical NLP services using ClinicalBERT/BioBERT, spaCy, PyTorch, Celery, and confidence thresholds with human validation. Deployed ML/API workloads using Kafka, Docker, Kubernetes, Jenkins, and monitoring with encryption and audit logging.