At Janea Systems, I build and maintain core components of an LLM-powered SQL assistant in production, from Qdrant retrieval and text-to-SQL generation to fine-tuned inference and multi-agent orchestration. Fine-tuning with LoRA and serving through vLLM and KServe helped reduce latency by 50%.
I also built evaluation and prompt-management frameworks, and an observability stack with Datadog and LangFuse. It tracks costs and production quality, contributing to a 30% reduction in inference costs.
Before Janea Systems, I designed data pipelines at Agile Actors on Databricks and Spark, processing more than 10 million events per day for a major Greek bank. I also led an on-premise-to-Azure migration and optimized Delta Lake queries.
At ICCS - NTUA, I trained and deployed forecasting and logistics models and worked on distributed deep learning research at the National Technical University of Athens. My PhD research produced ten peer-reviewed publications, and I’m now pursuing an MSc in Quantum Computing and Quantum Technologies.
