Khosrav Sinakmvi

Led creation and productionization of Document AI, LLM, and recommendation initiatives. Built an OCR + entity extraction pipeline using AWS Textract, LangChain, Terraform, Snowflake, and LLMs with deterministic fallbacks for reliable mapping to internal schemas. Developed golden evaluation datasets and integrated LLM regression tests into CI/CD to block releases when accuracy fell below thresholds. Implemented DataDog dashboards and custom LLM observability (latency, error, evaluation metrics). Built tenant embedding and reranker models using Snowflake data (including segmentation and recommendation use cases). Contributed production code for large-scale entity extraction and schema mapping to reduce manual review for high-volume document processing.

Developed an AI-enhanced industrial QA and asset management platform enabling co-pilot experiences for workflow management, audits, and monitoring across digital and physical assets. Designed serverless microservice architecture to decouple AI inference and training endpoints, integrating Node.js backend with React via REST APIs. Built chatbots trained on organizational data using RAG, retrieval/reranking mechanisms, Pinecone, and AWS OpenSearch; implemented multimodal retrieval with LlamaIndex/Neo4j/knowledge graphs and ontology-based retrieval. Developed object detection pipeline for video streams (vector embeddings via AWS Titan) and built AI agents to automate internal workflow audit generation and scheduling. Led a team of front-end and back-end developers and drove infrastructure-as-code using AWS CDK.

Built machine learning and ETL pipelines to analyze sensor data from PLCs and detect anomalies affecting product quality, enabling predictive maintenance and reducing annual scrap costs by approximately $500,000. Implemented anomaly detection pipelines with Flink, Azure File Storage, Azure Data Factory, Power BI, Azure Functions, and Docker; executed batch time-series analysis using TensorFlow 2.0, scikit-learn, NumPy, and pandas. Built ETL ingestion to bronze/silver/gold zones for anomaly detection, collaborating with IT and business intelligence teams for an end-to-end predictive maintenance architecture. Also delivered dashboards and implemented process improvements including factory-wide electronic 5S auditing and smart material requisition dashboards.