Victor Krim

Developed and maintained four simulated full-stack application environments and MCPs for training multimodal LLM agents, using React, Nest.js, Prisma, and Terraform. Performed manual UI, REST API, and WebSocket testing to ensure feature-spec adherence. Wrote end-to-end CI tests with Playwright achieving 100% coverage for key user journeys. Identified and applied optimizations to reduce Largest Contentful Paint by approximately 30%, saving significant agent compute time per training cycle.

Trained and deployed a two-way recommender using TensorFlow and Keras, increasing KPIs by 50–90% vs. random matching. Managed cloud-native distributed backend with GraphQL, PostgreSQL, and Kubernetes. Implemented session management with Redis, OAuth2, and JWTs. Enabled event-driven messaging and multimodal services using Kafka and AWS.

Developed ML data pipelines in Spark and Airflow for over 100,000 documents, equating to about 3 years of full-time effort. Implemented automated safety and quality metrics, achieving 100% sensitivity and 76% specificity, and addressed a security vulnerability related to model knowledge cutoff.