Sebastian D.

ML Engineer @ UserTesting

About

I’m a Machine Learning Engineer focused on developing reliable, interpretable, and high-impact AI systems. My current work revolves around Large Language Models (LLMs) — designing retrieval-augmented generation (RAG) architectures, implementing post-processing and validation layers, and developing guardrails against prompt injection, hallucinations, and other failure modes. Alongside my LLM work, I have a strong foundation in graph theory and traditional machine learning. I’ve built entity and hypergraph-based models to detect anomalies, extract semantic relationships, and improve downstream learning. I’m also interested in how graph neural networks (GNNs) and representation learning can enrich LLM systems — especially for reasoning, context retrieval, and structured data understanding. What drives me is the balance between rigor and creativity: I enjoy transforming complex, messy data problems into clear, documented, and reproducible solutions. Whether I’m engineering a new pipeline, exploring embeddings, or refining model evaluation metrics, I aim for both robustness and transparency. Outside of work, I’m based in Barcelona, where I balance my curiosity for AI with climbing, paragliding, and van trips through the mountains.

Country

Spain

City

Barcelona

Industry

Computer Software

Skill

AWS SageMaker, GNNs, Large Language Model Operations (LLMOps), AWS CloudFormation, Large Language Models (LLM), Amazon Web Services (AWS), Graph Theory, Interpersonal Skills, Predictive Analytics, Statistical Data Analysis, Excel Formulas, Natural Language Processing (NLP), Signal Processing, Vectorization, NumPy, Unstructured Data, Presentation Skills, Data Mining, Storytelling, Visualización de datos

Experience

ML Engineer

UserTesting

2024-1 - Present · 2 yrs 9 mos

Barcelona, Cataluña, España

Main Responsibilities - Design and implement LLM-powered Retrieval-Augmented Generation (RAG) systems that extract and synthesize insights from user testing data. - Develop evaluation frameworks to detect and reduce hallucinations, prompt injections, and other risks in generative models. - Build and maintain end-to-end ML pipelines in AWS SageMaker, including feature aggregation, embedding computation, and model deployment. - Apply graph theory and graph-based learning (Node2Vec, GNNs, hypergraphs) to improve entity relationships, anomaly detection, and knowledge extraction. - Collaborate with cross-functional teams to integrate ML solutions into production systems and ensure clarity, reliability, and scalability. Key Projects & Achievements - Designed a guardrail strategy combining AWS tools and custom logic to secure RAG applications against injection and hallucination attacks. - Built a postprocessing pipeline for entity extraction and categorization using graph-based co-occurrence networks and community detection. - Researched about a hyperedge anomaly detection framework to identify fraudulent participants in user test data. - Research on hallucination scoring model and other LLM metrics using semantic similarity computations to evaluate LLM responses. - Supported the InsightsDiscovery project, contributing to the integration of trustworthy AI and improving model interpretability and performance.

Data Scientist | AI Engineer

MINDBANK AI

2023-2 - 2023-12 · 11 mos

Prague, Czechia

• Design and implement algorithm for the creation of Digital Twins, from user data collection to developing conversational agents using SOTA NLP technology. • Resulted in contracts with 5 major clients and with at least 3 potential investors. • Employed quantization to deploy open source LLMs hosted within the Hugging Face transformers library in company servers, cutting third party expenses by (potentially) 60%. • Develop custom algorithm for bio signal processing, using numpy and pandas to extract HRV measurements from video recordings. Primary research with open dataset translated in collaboration with 2 big universities.

Game designer

KobGames LTD

2021-12 - 2023-6 · 1 yr 7 mos

• Lead development teams for hyper-casual games, focusing on user engagement and retention strategies. Used data-driven insights to optimize game mechanics. • Conducted rapid A/B testing using player data to inform design decisions, resulting in improvements up to 15% in some KPIs (CPI and D7 retention). • Collaborated with cross-functional teams to bring creative concepts to production. Demonstrated adaptability in navigating the dynamic and rapidly evolving landscape of the hypercasual games industry. • Developed pluggable Python and C# scripts to track game economy metrics in the games, both during development and in published games, gathering important data for decision making.