Aarush Dubey

Fair Conformal Classification for Financial Transactions (AISTATS 2027)

Solved the coverage–equity paradox in high-stakes financial ML by engineering a dual-objective calibration algorithm balancing group coverage and prediction set-size parity. Formulated and proved Theorem 1 (worst-group coverage lower bound). Cut set-size disparity by 34% with 100% regulatory auditability (FCRA, ECOA, EU AI Act compliant).

Cloud-Deployed RAG Document Intelligence Engine

3-tier RAG pipeline for intelligent multi-document querying with NVIDIA AI Foundry. Custom C++ parser achieves 7.5x speedup. Serves 300+ users, 5,000+ queries processed.

Real-Time Computer Vision Analytics Platform

Production-grade platform integrating YOLOv8 object detection and Optical Flow analysis with FastAPI backend. Processes 30 fps with 10,000+ logged detection events.

AI-Powered Financial Intelligence Twin

AI guide leveraging NVIDIA Llama 3.1 (70B) and Groq for hyper-personalized financial insights. 150+ registered users, sub-1s page transitions, 98% uptime.

Explainable ML Pipeline for Financial Intelligence

End-to-end ML pipeline for automated financial transaction categorization. Achieved macro F1-score of 1.00 using ensemble methods with SHAP-based explainability.

Real-Time Cybersecurity Threat Detection Platform

Browser extension with NLP-based threat detection to identify phishing, social engineering, and malicious content in real-time. 94% detection accuracy.

Autonomous AI Agent with Long-Term Memory

Sophisticated AI Personal Assistant with persistent memory using JSON-based memory systems. Integrated Google Gemini API for real-time task automation and planning.