Manolis Papageorgiou

Lead GenAI development in Python with LangGraph, LangChain, and MCP, including a multi-agent system (planner, research, writer, and reviewer agents) that drafts Horizon Europe proposals aligned to call requirements. Build RAG pipelines with layout-aware parsing (Azure Document Intelligence), Azure OpenAI embeddings, and hybrid search in Azure AI Search. Fine-tune local LLMs with LoRA/QLoRA for private, on-premise chatbots. Deploy FastAPI services on Azure with Docker and Helm; trace and evaluate in Langfuse.
Core stack: Python, LangGraph, LangChain, MCP, RAG, LLM fine-tuning, FastAPI, Langfuse, Azure OpenAI, Azure AI Search, Docker.

Developed machine learning models for early detection of cognitive disorders (MCI, dementia, Alzheimer's) from voice recordings, from Random Forest and XGBoost to a multimodal TensorFlow/Keras neural network fusing speech, text (ASR transcripts) and demographic features. The network outperformed single-modality and tree-based baselines under patient-level cross-validation. Used SHAP to explain predictions and validated outputs with clinical experts.

Built legal document AI: fine-tuned BERT-based models for clause classification, an EfficientNet CNN for page deskew regression, and YOLO on our labeled data to detect tables, images, headers/footers.

Built deep learning NLP models (word embeddings + BiLSTM) on job posts and CVs for job classification and siamese candidate–job matching, and automated CI/CD pipelines to deploy ML models.

Member of the task force working on project "Tellody", a web platform created for a major telecommunications provider.
Technologies used: Java, J2EE, spring, maven, JBOSS, jsp, MySQL, MongoDB