At SoftwareMill, I lead client-facing AI engagements from discovery and feasibility assessment through solution architecture and PoC delivery. In a current Document AI engagement, I designed an asynchronous document-processing architecture on GCP and evaluated OCR, object detection, segmentation, and VLM approaches against quality, latency, and cost.
I independently developed and deployed an ML-based payment-to-invoice reconciliation system, including its GCP infrastructure and automated testing, retraining, and deployment pipelines. I’ve also worked on production robotic vision monitoring and improvements, investigating model failures and coordinating annotation, fine-tuning, and deployments.
Earlier at VoiceLab, I researched multilingual text generation and developed NLP components for meeting-note automation and chatbots. I co-authored research that led to a publicly available model, and co-authored Transformer-based NLP research.
