Understanding complex systems through adaptive intelligence.

Research profile

Four connected research pillars

My work combines adaptive agents, statistical verification, robust quantitative methods, and large-scale text analysis to study complex economic and social systems, from simulation design to empirical validation.

1. Adaptive Multi-Agent Systems

   I study populations of learning, heuristic, and deliberative agents interacting over evolving economic networks, studying how local adaptation shapes aggregate dynamics.

   Multi-Agent Systems · Reinforcement Learning · Graph Neural Networks

   Expand details & projects

   In this research line, I model economic systems as dynamic, decentralized networks of heterogeneous decision-makers. Rather than assuming hyper-rational global equilibrium, agents operate with bounded rationality, local interactions, and adaptive strategies (such as multi-agent reinforcement learning or heuristic decision rules). By combining network science and agent-based modeling, I analyze how micro-level adaptation gives rise to macroeconomic phenomena, market liquidity shifts, and emergent systemic patterns.

   Relevant Projects

   RiskSentinel - Agentic Systemic Risk Simulator Multi-agent systemic risk simulator for financial contagion analysis on 210 S&P 500 stocks, built for Microsoft AI Dev Days Hackathon 2026 and deployed as an interactive Streamlit app. View project Multi-Agent Orchestration Event-driven multi-agent coordination system for auction-based procurement, inventory allocation, and real-time fulfillment under demand uncertainty. View project Real Estate AI Agent Autonomous AI agent for property-market analysis, price estimation, and natural-language interaction built on predictive models and LLM orchestration. View project

2. Statistical Verification

   I use convergence diagnostics and statistical model checking to make simulation evidence more reliable and reproducible.

   Statistical Model Checking · Agent-Based Modeling · Reproducibility

   Expand details & projects

   Agent-based models and complex stochastic simulations generate massive execution traces that are notoriously challenging to validate formally. My work applies Statistical Model Checking (SMC), hypothesis testing, and rigorous statistical convergence diagnostics to verify complex simulation outcomes against temporal logic specifications. This bridge between formal methods and agent-based economics establishes quantifiable confidence intervals for policy interventions and stress-test scenarios.

   Relevant Projects

   Island Model + MultiVeStA — Statistical Model Checking of Economic Growth Accepted paper at MARS @ ETAPS 2026. Reproduces and extends Fagiolo & Dosi (2003) Island Model using MultiVeStA for statistical model checking, with formal convergence guarantees on parameter sensitivity analysis. View project

3. Robust Quantitative Methods

   I apply econometrics, robust statistics, and graph learning to systemic risk and financial decision problems.

   Econometrics · Systemic Risk · Economic Networks

   Expand details & projects

   To address instability and structural shocks in economic infrastructure, I integrate robust econometrics, extreme value theory, and graph neural networks. This approach enables early detection of financial contagion vectors, stress propagation in interbank market networks, and energy network vulnerability forecasting. The goal is to build quantitative decision-support frameworks that remain resilient under heavy-tailed distributions and structural market regime shifts.

   Relevant Projects

   Robust Portfolio Optimization under Systematic Market Disruptions (PFSE) Novel PFSE estimator achieves 25% breakdown point with 32× computational speedup over MCD. Out-of-sample Sharpe 1.87 on S&P 500 (2015–2025), 29% lower drawdown during COVID-19. Submitted to Computational Economics. View project Network Topology Analysis for Systemic Risk Prediction Financial machine learning project combining dynamic correlation networks, graph neural networks, and trading backtests to detect systemic risk in U.S. equity markets. View project Gas Network Risk Forecasting Second-place hackathon project for gas leak risk prediction using geospatial-temporal features, synthetic data augmentation, and SHAP-based interpretability. View project

4. Text Analytics and Language Models

   I apply text analytics and language models to multilingual corpora, sentiment and narrative dynamics, and to systems based on retrieval and embeddings.

   NLP · Narrative Economics · Foundation Models

   Expand details & projects

   I apply text analytics methods and language models to the study of unstructured data. The work covers the semantic structure of multilingual corpora, the extraction of sentiment and narrative dynamics, and the construction of systems based on retrieval, custom embeddings, and parameter-efficient adaptation of pretrained models. I use classical text analytics alongside transformer architectures, state-space models, and foundation models, with attention to verifying that results depend on the data rather than on the technique employed.

   Relevant Projects

   NLP & Semantic Network Analysis Technical implementation of the JADT paper pipeline, integrating multilingual preprocessing, topic modeling, correspondence analysis, semantic spaces, sentiment modules, and network-based validation across methods. View project Lightweight Fine-Tuning with PEFT & LoRA Parameter-efficient fine-tuning pipeline using LoRA to adapt DistilBERT for sentiment analysis while training less than 1% of the model parameters. View project Custom Chatbot with RAG Retrieval-augmented chatbot that combines semantic search, custom embeddings, and LLM prompting for grounded domain-specific conversations. View project