ARGOS, ATLAS & Odyssea
HPC Web Servers for Medical & Industrial AI
Problem
SISSA's cutting-edge mathematical models lived in Jupyter notebooks with no web interface | inaccessible to non-technical researchers and external collaborators worldwide.
Solution
Architected three production web servers (ARGOS, ATLAS, Odyssea) wrapping Python simulation backends in React UIs with real-time result streaming via WebSockets.
Result
Researchers across 15+ countries now access complex simulations through a browser. FAST Computing startup launched with full web infrastructure.
Impact
- ✦3 production web servers deployed and maintained
- ✦Researchers across 15+ countries served
- ✦Complex mathematical simulations made accessible via web UI
- ✦FAST Computing startup infrastructure built from scratch
01The Problem
SISSA's Mathlab had cutting-edge mathematical models for medical and industrial AI simulations | but they lived entirely in Jupyter notebooks and command-line scripts. Researchers outside the lab couldn't access them. The models were too complex for non-technical users. There was no web interface, no API, no deployment infrastructure. The gap between the research and its real-world application was enormous.
02My Role
As the sole Research Fellow responsible for the web infrastructure, I was tasked with architecting three independent computational web servers | ARGOS (focused on medical simulations), ATLAS (industrial AI), and Odyssea (data visualization and analytics). Each needed to wrap complex Python simulation backends in intuitive, accessible web interfaces. Simultaneously, I built the complete web infrastructure for FAST Computing, a SISSA-born startup commercializing this research.
03Technical Approach
Each server followed a similar architecture: a Django or Flask backend exposing the simulation models as REST APIs, a React frontend providing interactive parameter controls and result visualization, and Plotly/Dash for real-time data visualization. The simulation models themselves were written in Python with heavy NumPy and SciPy usage. The biggest challenge was performance | mathematical simulations that could run for minutes needed to feel responsive to users. I implemented async job queuing with real-time progress updates, caching for repeated parameter sets, and WebSocket connections for live result streaming.
04Key Decisions
Choosing Django over FastAPI gave us the admin interface for free | researchers could manage simulation parameters and results without touching the codebase. React was chosen for the frontend because the team had some prior exposure. Plotly Dash handled the visualization layer because it integrated directly with the Python simulation outputs without needing a serialization layer. For FAST Computing, I made the decision to use a headless CMS architecture so the scientific team could update content without developer involvement.
05Outcome
All three servers went live and are actively used by researchers across Europe and beyond. FAST Computing's website successfully launched the startup's public presence. The work directly contributed to bridging the gap between SISSA's theoretical research and real-world industrial application | the core mission of the Mathlab group.