Victor Barbuta

Developed a multi-modal deepfake detection pipeline using Vision Transformers, Wav2Vec2, and CLIP embeddings; trained adversarial detectors with PyTorch Lightning + AWS SageMaker; implemented real-time feature extraction via OpenCV, FFmpeg, and TensorRT; designed a hybrid CNN/Transformer ensemble achieving 97.9% accuracy on in-the-wild deepfake benchmarks; automated retraining with Vertex AI Pipelines and MLflow; built predictive ML pipelines for retail and e-commerce demand forecasting; ETL via Google Dataflow and BigQuery; dashboards in Tableau; API-driven automation with Zapier/n8n/Make; deployed inference APIs as microservices.

Architected scalable graph-based ML pipelines for entity resolution, fraud detection, and network anomaly detection; implemented real-time risk scoring; built multi-source ETL pipelines with Apache Spark, PySpark, Airflow, BigQuery, and GCP Dataflow; developed MLOps workflows with Docker, Kubernetes, Terraform, Helm, GitHub Actions, Argo Workflows; applied GNNs, XGBoost, LSTM/RNNs, and NLP transformers for relationship inference, anomaly detection, and textual data analysis; monitored pipeline health and model drift with Prometheus, Grafana, Datadog, ELK; optimized real-time inference with FastAPI, gRPC, Redis, Kafka; exposed secure, API-first ML endpoints for SaaS platform.

Designed and implemented RESTful APIs and microservices powering high-traffic web and mobile applications across e-commerce, fintech, and healthcare; built backend architectures using Python, Node.js, and PHP frameworks; optimized relational and NoSQL databases; collaborated with front-end, mobile, and QA teams; integrated third-party APIs; implemented CI/CD pipelines and automated testing; mentored junior developers and contributed to cloud-native architecture modernization.