Trustworthiness
Evidence strength, uncertainty, provenance, and limitations remain visible throughout the reasoning process.
Peer-reviewed research
Scientific work for trustworthy decisions.
Research in Bayesian machine learning, uncertainty quantification, evidence-governed AI, healthcare, autonomous systems, fraud detection, decision intelligence, and optimization.
Research standard
Rigor for high-stakes environments
TeraSystemsAI research is organized around measurable reliability, explicit assumptions, reproducible methods, and disciplined limits on what an intelligent system may claim or decide.
Efficiency
Methods are evaluated not only for accuracy, but also for operational usefulness, latency, and deployment constraints.
Reliability
Robustness, calibration, failure modes, and degraded conditions are treated as central scientific questions.
Accountability
Human oversight, traceability, abstention, and responsibility boundaries are built into the research design.
Publication record
Selected peer-reviewed publications
Each paper is presented with its publication record, core scientific contribution, and direct DOI link. Promotional view counts and unverifiable impact claims are intentionally excluded.
2026June
BRAG: Bayesian Retrieval-Augmented Generation; A Methodological Framework for Evidence-Governed Decision Support
Machine Learning and Knowledge Extraction, 8(6):151
Contribution. Introduces a Bayesian, evidence-governed framework that quantifies epistemic uncertainty over retrieved evidence and carries that uncertainty into generation, enabling systems to qualify, defer, or abstain when support is insufficient.
Governance
Conclusions remain bound to evidence strength and provenance.
Uncertainty
Epistemic uncertainty is surfaced rather than hidden behind fluent output.
Method
A reusable framework for trustworthy retrieval-augmented decision support.
2026March
Stochastic Inventory Optimization with Coherent Risk Measures: A Decision-Theoretic Framework for Probabilistic Forecasting and Constrained Optimization
Journal of Risk and Financial Management, 19(3):173
Contribution. Unifies probabilistic demand forecasting, constrained newsvendor optimization, and coherent tail-risk evaluation through Conditional Value-at-Risk, linking calibrated quantile forecasts to operational decisions.
Forecasting
Quantile regression connects predictive distributions to order decisions.
Risk
CVaR evaluates adverse tail outcomes rather than average performance alone.
Operations
An end-to-end framework bridges prediction and constrained action.
2026January
Bayesian RAG: Uncertainty-Aware Retrieval for Reliable Financial Question Answering
Frontiers in Artificial Intelligence, Volume 8
Contribution. Integrates epistemic uncertainty directly into retrieval through Monte Carlo Dropout and a risk-sensitive scoring function that balances semantic relevance against uncertainty in financial question answering.
Retrieval
Ranks evidence by relevance while penalizing uncertain representations.
Calibration
Evaluates whether confidence aligns with observed answer reliability.
Deployment
Designed for modular integration into existing RAG pipelines.
2025May
Enhancing Autonomous Systems with Bayesian Neural Networks: A Probabilistic Framework
Frontiers in Built Environment, Volume 11
Contribution. Applies Bayesian neural networks to autonomous navigation so that predictive uncertainty can inform more conservative behavior under noisy sensing, changing weather, and unfamiliar traffic conditions.
Navigation
Uncertainty identifies conditions where deterministic confidence is unsafe.
Resilience
Probabilistic outputs support cautious behavior under degraded sensing.
Oversight
Decision confidence can be inspected and incorporated into safety controls.
2025May
Hybrid Naïve Bayes Models for Scam Detection: Comparative Insights From Email and Financial Fraud
IEEE Access, Volume 13
Contribution. Compares interpretable Naïve Bayes methods with hybrid deep-learning and ensemble architectures across email phishing and highly imbalanced financial-fraud datasets.
Email
Evaluates detection of subtle phishing patterns in large email collections.
Transactions
Examines extreme class imbalance in credit-card fraud detection.
Interpretability
Balances predictive performance with models that remain understandable.
2024November
Application of Bayesian Neural Networks in Healthcare: Three Case Studies
Machine Learning and Knowledge Extraction, 6(4)
Contribution. Examines Bayesian neural networks across three healthcare cases to demonstrate how predictive performance and uncertainty estimates can be considered together in high-stakes clinical modeling.
Diabetes
Studies personalized prediction for treatment and HbA1c outcomes.
Alzheimer's
Explores early detection using clinical and genetic information.
Uncertainty
Compares probabilistic confidence with standard predictive approaches.
2024September
A Comparative Study of Optimization Techniques on the Rosenbrock Function
Open Journal of Optimization, Volume 13
Contribution. Benchmarks major optimization methods on the non-convex Rosenbrock function, comparing convergence behavior, stability, and gradient dynamics across difficult optimization landscapes.
Benchmark
Uses a difficult curved valley to expose optimizer behavior.
Comparison
Contrasts gradient descent, momentum, and adaptive methods.
Insight
Clarifies tradeoffs in convergence speed and stability.
Research progression
A coherent publication program
The record develops from optimization and Bayesian healthcare modeling toward uncertainty-aware retrieval, autonomous systems, fraud detection, decision intelligence, and evidence-governed AI.
2026
Evidence-governed generation and decision intelligence
BRAG, Bayesian RAG, and stochastic inventory optimization connect uncertainty, evidence, risk, and operational action.
2025
Safety, autonomy, and fraud detection
Research expands Bayesian reasoning and hybrid models into autonomous navigation, email scams, and financial fraud.
2024
Foundations in healthcare uncertainty and optimization
Bayesian neural networks and optimization benchmarks establish the methodological base for later high-stakes systems work.
Research areas
Where the work is concentrated
These areas reflect the themes represented in the publication record and TeraSystemsAI's broader research and engineering program.
Bayesian Machine Learning
Trustworthy Artificial Intelligence
Uncertainty Quantification
Evidence-Governed AI
Healthcare AI
Decision Intelligence
Autonomous Systems
Optimization and Risk
Research collaboration
Discuss Collaboration
Advance trustworthy AI with scientific discipline.
TeraSystemsAI works with researchers, institutions, public-interest organizations, and industry teams on uncertainty-aware systems, evidence governance, evaluation, and high-stakes decision support.