Eirini Stamatiou

NLP and GenAI engineer with 6+ years' experience at Shell building production LLM systems. Recent work includes a contract intelligence platform combining agentic workflows and RAG — capturing ~$1.5M in value — and a customer-operations automation tool deployed across European markets, saving ~9 FTE.

Comfortable across the full applied AI stack: Python, Azure (AI Foundry, AI Search), FastAPI, RAG, prompt engineering, and LLM evaluation. Looking for AI Engineer roles where I can take LLM-powered systems from concept to production in real business contexts.

NLP Data Scientist

Shell · London, UK · Hybrid · 2019 – Present The Beast AI — Contract Intelligence Platform · 2026 – Present

Engineer on a cross-functional team on an LLM-powered platform that processes ~7,500 contracts; the platform has replaced ~4 months of manual analyst work and captured ~$1.5M in value to date.

Contributed across a two-stage architecture inspired by Karpathy's LLM-wiki pattern: extractors that parse contracts into structured markdown, and a tool-using agent that consolidates per-customer information into a "customer wiki," with tools for search, retrieval, and Salesforce integration.

Worked on the RAG layer using Azure AI Search (hybrid vector + keyword retrieval), powering natural-language Q&A over contract content.

Contributed to the FastAPI backend serving the platform, working within an existing microservice architecture.

Deep Research Agent — Internal Research Assistant · 2025

Built the production web backend for an internal agentic RAG application for researchers: FastAPI service handling authentication, user management, rate limiting, and conversation persistence.

Customer Operations Email Automation · 2023 – 2024

Core contributor to an LLM-powered tool embedded in Salesforce Service Cloud that automates email handling for Customer Operations. Initially rolled out for the Lubricants business across European markets, using intent recognition to detect order-placement requests and trigger structured order data extraction. Subsequently extended to Business Mobility, where the same platform was repurposed for customer verification.

Worked across the full stack: intent recognition, structured data extraction from email bodies and attachments, prompt engineering for the underlying LLM (GPT-3.5), and the Python orchestration layer mediating between the LLM, Salesforce, and Shell's ERP system.

Implemented mitigations for hallucination risk, including handling of ambiguous customer inputs (units of measure, regional number formats) and surfacing all proposals for verification before order placement or email send.

Set up evaluation to catch regressions and improve prompt/extraction quality across releases. Delivered ~9 FTE saved in Business Mobility and 5–8 minutes saved per order.

MSc in Speech and Language Processing (Merit) · University of Edinburgh, UK · 2018 – 2019

BSc in Linguistics (Distinction) · National and Kapodistrian University of Athens, Greece · 2012 – 2017