At Reply, I design and ship multi-agent workflows in Python, using tool calling, retrieval, and context engineering to move AI applications from proof of concept to production. I use typed Pydantic contracts and structured outputs to make model responses more predictable for downstream systems.
I evaluate agent output quality, retrieval correctness, indexed content, and integration calls, and debug issues across prompts, orchestration, and backend services. My work also includes Elasticsearch and MongoDB data layers, and deploying approved solutions through Docker and Jenkins CI/CD on Google Cloud Run.
Outside work, I build consumer AI products end to end. For MooWeather, I created an evaluation harness of 600+ test calls to measure structured-output reliability across models; I also built World News Monitor and HeyPantry!. My research at the University of Bergamo examined linguistic analogies in word embeddings, and I hold a Master’s Degree in Computer Engineering with a final grade of 110/110.
