Mirco Gnuva — I build LLM-powered agents and tools in Python: from autonomous browser agents to self-evaluating RAG pipelines.
Single-page Astro static site with a dual light/dark 'blend' design and an animated canvas background. No UI framework; deployed on Cloudflare Pages with automatic builds on push.
Vision LLM agent that plays OpenGuessr autonomously: drives a real browser via Playwright, reasons over panorama screenshots, and runs a two-stage exploration → guessing loop with Langfuse tracing.
Multimodal pipeline that ingests PDFs, scanned documents, images, tables, and audio (via faster-whisper), then generates study notes with an LLM and auto-evaluates them through a self-correcting regeneration loop.
Multi-agent meal-planning assistant (master + plan-generator agents) with a multilingual Chainlit UI, eval-golds dataset for quality measurement, and full Docker Compose deployment.
Conversational packing assistant powered by a pydantic-ai agent with persistent mem0 memory, a Chainlit UI, FastAPI backend, SSE streaming, and Langfuse observability.
Shared movie backlog and multi-person voting app: each member rates candidates, the app surfaces the consensus pick. Streamlit UI, FastAPI backend, MongoDB storage.
I'm an AI Engineer who thinks in full cycles — from architectural design and stakeholder alignment to implementation, evaluation, and long-term performance monitoring. Shipping a feature is the beginning, not the end.
At Elmec Informatica I build LLM agents, RAG pipelines, and evaluation infrastructure. The decisions I care most about are the ones that determine whether a model-powered feature is durable: the eval harness, the feedback loops, the observability layer that tells you when something quietly degrades.
Outside work I run personal experiments and follow research to stay ahead of what's possible. I'm methodical enough to want systems that are measurable and honest about their failure modes — and curious enough to explore ideas before they're proven.
Lead the retrieval and evaluation platform powering the core product. Set the bar for how models are tested before they reach users.
Developed software solutions integrating different tools to create new products and services.
Laurea Magistrale in Scienze e Tecnologie Informatiche.
Laurea triennale in Informatica. Graduated 100/110.
Elettronica e Elettrotecnica. Graduated with top marks.