Vasileios – Python, LLM, AI agent development

Build an agentic customer support system that can handle new orders, product-related questions, discount requests, shipping issues, and damaged product claims.

• Led the architecture and development of a multi-agent AI orchestrator to automate customer and merchant support workflows;
• Designed the system architecture and built the initial MVP end-to-end;
• Implemented the core orchestration logic and service layer with FastAPI;
• Containerized services with Docker and implemented the CI/CD and deployment pipeline;
• Published images to AWS ECR and ran scalable services on AWS ECS behind an Application Load Balancer;
• Managed and mentored a team of 10 AI Engineers;
• Architected and built the PoC for a media-buying automation tool integrated with the orchestrator to optimize ad spend and campaign performance in real time.

Designed and implemented an AI-powered digitization system that transformed unstructured pharmaceutical manufacturing recipes from PDFs into structured knowledge graphs using LLMs, RAG, GraphRAG, and graph databases.

• Designed and developed AI solutions to digitize batch processes and automate document processing workflows;
• Implemented a hybrid search system enabling intelligent retrieval across structured and unstructured data;
• Developed a Python-based search engine locally and exposed its functionality through FastAPI services;
• Containerized the application using Docker and deployed images to AWS ECR;
• Ran scalable services on AWS ECS behind an Application Load Balancer;
• Integrated Amazon Aurora for structured data and Amazon S3 for storage and data access within the service architecture;
• Leveraged LLMs, RAG pipelines, and AWS cloud services to build scalable and high-performance ML solutions for enterprise applications;
• Decreased batch record import time from 3 months to 1 week;
• Improved batch processing quality by eliminating manual errors through automation.

Developed machine learning models using Random Forest, XGBoost, and CatBoost to predict auction repetition rates and redefault probability, supporting debt recovery optimization and portfolio management decisions.

• Predicted auction repetition number and redefault probability with varied algorithms;
• Conducted data analysis for non-performing loans and auctions to assist senior management in devising action plans;
• Automated regulatory reporting procedures for European and Greek market and bank authorities;
• Facilitated revenue increase from $400K to $600K per week by applying machine learning algorithms to reduce the auction cycle time of real estate by 50%;
• Steered the reporting process automation aimed at swiftly presenting critical insights to regulatory authorities for loans worth $30B;
• Decreased regulatory report creation time by 10x from 100 to 10 man-days;
• Mitigated regulatory penalty risks through elimination of manual errors.

Contributed to a three-person team that designed and built an enterprise-scale data lake for a leading retail bank, enabling centralized data management and analytics across the organization.

• Engaged in the company’s data lake development process to secure structured, unstructured, and semi-structured information;
• Automated ETL data flow for day-to-day updating procedures;
• Visualized data in line with analytical queries to support ad-hoc reporting;
• Augmented data fields using PySpark, Python, SQL, Cloudera Stack CDH, Hive, Impala, Airflow, Oozie, D3.js, Alteryx, Dash, and Presto;
• Transitioned data analytics from relational to NoSQL databases using the Cloudera framework;
• Built and scaled up a data lake from 0 bytes to 1.5 PB;
• Facilitated the creation of the Personal Financial Management (PFM) banking product through deep data analysis and data flow design and implementation.

Performed RFM analysis and statistical modeling on customer and advertising data to generate actionable insights, improve targeting strategies, and support data-driven business decisions.

• Established the Business Intelligence and Analytics Department to gather and store data from several sources;
• Conducted regression analysis to search for correlation between several variable parameters;
• Coordinated customer segmentation procedures such as RFM, cluster analysis, and RFE;
• Prepared and presented reporting results using PostgreSQL and R;
• Improved customer lifetime value by 20% through customer segmentation and engagement initiatives, including recency frequency monetary value analysis.

Built and optimized game engines for Video Lottery Terminal (VLT) slot machines, contributing to core gameplay logic, performance, and regulatory-compliant gaming operations.

• Defined slot machine reels playability and forecasted short- and long-term profits to support executive decision-making;
• Implemented the company’s validation testing framework using SoapUI;
• Achieved GLI certification required for slot machines’ marketability.