Dmitry K.

RAG-enhanced voice assistant for HR support & psycho-linguistic screening with strict compliance. Collaborated with Legal on AI risk tolerance. Built multi-step LangGraph RAG architecture with LangChain LLM chains & advanced prompts. Optimized Qdrant indexing for millions of legal texts. Deployed Llama LLMs via Triton/vLLM on CPU/GPU. Tuned SentenceTransformers for finance/legal embeddings. Established Ragas compliance framework (Faithfulness/Groundedness). MLflow tracked experiments/models. Curated multi-domain test corpus. Added "no-hallucination" controls & policy prioritization. AWS monitoring (Prometheus/Grafana/OpenTelemetry). Env: Python, PyTorch, Hugging Face, LangChain/Graph, MLflow, Qdrant, Triton/vLLM, Kubernetes, AWS EKS/Sagemaker.

Machine Translation platform for secure on-premise translation across 109+ languages, documents, websites. Supported custom terminology, audio recognition (91 langs), subtitles, diarization. Achieved BLEU/COMET scores rivaling DeepL/Google. Designed Whisper fine-tuning pipeline with ClearML/AWS S3. Containerized microservices (Docker) on AWS ECS/Fargate. Built complete data prep/training pipeline for MT models. Novel Unicode/multilingual methodology improved emoji/foreign text quality. Structural/semantic filtering boosted training datasets. Fine-tuned Gemma2-9b for post-processing. Researched data filtering/model optimization. Authored 90+ page MLOps docs. Collaborated with linguists, mentored engineers. Env: Python/TS, TensorFlow/PyTorch, Hugging Face, OpenNMT, MarianMT, NLLB, Docker, AWS EKS/Sagemaker, ClearML, FastAPI

Scalable Big Data platform/DWH for tele-veterinary subscription system processing payments, chat, demographics for BI/predictive modeling via Hadoop/Spark ML. Analyzed tariff purchases, client acquisition, service quality. Designed dimensional DWH schema in Greenplum MPP. Built Airflow DAGs for batch/streaming ETL. Used SparkKubernetesOperator for PySpark jobs on K8s. Optimized PySpark ETL (cleaning, transformation, features) loading to DWH. Managed S3 Data Lake (Parquet partitioning). Implemented Kafka streaming pipeline with Spark Streaming. Set up MLflow Tracking (MinIO/PostgreSQL) for Spark ML experiments. Containerized Spark apps (Docker/K8s) with monitoring/profiling. SQL quality checks via Airflow operators. Env: Python, Airflow, Spark/PySpark, Hadoop, Kafka, Greenplum, MLflow, Docker/K8s, GitLab CI, MinIO.