Adversarial games. Adaptive systems.
Quantitative research across reinforcement learning, game theory, and systematic finance.
Sri Atluri.
Formulating novel game-theoretic methods for computer vision. The current work designs multi-agent reinforcement learning games where adversarial agents learn to localise anomalies in images without supervision: no labels, no domain knowledge, no prior about what the anomaly looks like. The frameworks span anomaly detection, medical imaging, and frontier model evaluation through LLM-driven tool use.
Before RL, the work was in quantitative systems. A background in mathematics, statistics, and financial engineering led to Tevvis: a systematic equity strategy deployed on real capital since 2023, and still running. That same instinct for building adaptive systems under uncertainty carries through to the current research.
Currently a CRUK-funded PhD student in Computer Science at Queen Mary University of London, supervised by Dr Shaheer Saeed.
Projects and publications.
Tevvis: Systematic Equity
A four-stage Bayesian framework for capturing extreme-value-creating assets from heavy-tailed return distributions. Live on real capital since January 2023. The name means 23 in Marathi.
Bayesian Stock Screening and Cluster-Based Portfolio Optimisation in Distressed Markets
Develops a method for harvesting Bessembinder wealth: the empirical finding that a small number of stocks account for all net wealth creation in equity markets. Proposes a Bayesian screening stage followed by cluster-based portfolio construction that avoids the instabilities of classical mean-variance optimisation, tested on distressed-market regimes.
MICCAI Workshop Paper
Co-authored peer-reviewed paper accepted at a MICCAI workshop. Research at the intersection of machine learning and medical imaging.
Features for Screening Versus Features for Clustering: A Methodological Distinction in Systematic Equity Investment
Short-Horizon Excess Returns in Liquid Equities: Regime-Dependent Properties of a Systematic Programme
Return Predictability or Risk Timing? Bootstrap Evidence from a Century of US Equity Data
The CDS-Bond Basis: A Markov-Switching Friction Decomposition with Nonlinear Amplification
Currency Factor Combination and Cross-Sectional Pricing: Shrinkage Convergence, Alternative Methods, and the Identification of Volatility Risk
Clustering-Based Alternatives to Mean-Variance Portfolio Optimisation
Probability and Machine Learning Methods in Predictive Modelling: Applications to Alternative Data
Systematic equity.
A four-stage Bayesian framework for capturing extreme-value-creating assets from heavy-tailed distributions.
Performance represents live, real-capital results from a funded portfolio managed since January 2023. Past performance is not indicative of future results. Tevvis is a research operation, not an investment adviser.