My name is Mattia Redaelli and I am Senior AI Engineer @Miro. I specialize in building scalable LLM systems, agentic evaluation infrastructure, and multi-agent orchestration (MCP, A2A). I've delivered production-grade AI platforms from architecture through deployment, including a near-real time observability and tracing system for Miro's AI harness, and an AI Gateway optimizing LLM routing for cost, latency, and throughput. I've also worked on GraphRAG solutions leveraging knowledge graphs and ontologies for complex, knowledge-intensive reasoning, and on MCP/A2A registries to enable standardized agent interactions across teams.

• Agentic AI Evaluation Infra: built a near-real time observability and tracing system for Miro's AI harness, processing 1.14M+ events weekly, using Kotlin, Kafka, Braintrust, and Google DLP.

• Enterprise AI Gateway & Multi-Agent Platform (Retail): designed and developed a centralized AI Gateway to orchestrate LLM calls and monitor cost, latency, and throughput; led architecture definition and standardization of MCP/A2A multi-agent workflows, working directly with client stakeholders.
• Knowledge Graphs & GraphRAG (Retail): led an R&D initiative on graph-based RAG, integrating knowledge graphs and ontologies for knowledge-intensive applications, evaluating prototypes and use cases for a product serving 500k monthly active users.

• AI Assistants & RAG Platform (Transportation): designed and built an end-to-end RAG platform with tool-enabled AI agents for file handling, action execution, and context-augmented document QA, using OpenAI APIs and Azure; led cloud-native API development and integrations.
• Developed a comprehensive Data Platform on Snowflake with Data Mesh principles; used Kafka for real-time streaming and Terraform for infrastructure management.

• Led the design and maintenance of a scalable, robust front-end architecture, ensuring seamless integration with backend systems and effective collaboration within an Agile methodology.

Thesis: Enhancing Kafka load-balancing with consumer groups

Competition of the course of Recommender Systems at the Polytechnic University of Milan. Competed with other students to create a model for recommending new tv series to users of a given dataset.

The aim of the project is to predict future samples of a multivariate time series. Project developed during the course of Artificial Neural Networks and Deep Learning at the Polytechnic University of Milan.

The task was to develop a model for classifying images and to train it to distinguish between 14 classes of leaves. Project developed during the course of Artificial Neural Networks and Deep Learning at the Polytechnic University of Milan.

Software Engineering Final Project - BSc thesis. Modeling a board game as software, enabling remotely played games using sockets, and implementing all game rules. Playable both from CLI and GUI. Developed using Java, and JavaFX for the GUI. Implemented using Model View Controller architectural pattern (MVC).

Algorithms and data structures project. The project consists in a text editor written in C, with the possibility of performing multiple undo and redo, during insertions, as well as the usual printing and deleting.

• Italian – Native
• English – Professional Proficiency (EF SET C2 Proficient, TOEIC 950)

Passionate about cooking, running, Formula 1, board games, cinema, and music.