I'm a PhD candidate in Aerospace Engineering at the University of Texas at Austin (graduating in May 2027), where my research lives at the intersection of graph-based machine learning, reinforcement learning, and autonomous AI systems applied to large-scale, adversarial real-world problems.

My dissertation work centers on ASTRIAGraph — a production-scale heterogeneous knowledge graph of satellites, space operators, and spectrum allocations — where I designed and deployed an RL-based anomaly detection system to autonomously identify behavioral compliance violations among adversarial actors.

🏛️ 2026 Summer Graduate Archer Fellow · University of Texas at Austin
📄 Published: AMOS Conference 2025 · Acta Astronautica · US Space Force Symposium 2025
🎯 Research interests: Graph neural networks · Agentic AI systems · Anomaly detection · LLM orchestration

Neo4j · PyTorch Geometric · Python · GDS

Built a heterogeneous transaction graph (users, devices, IPs, merchants) and applied Weakly Connected Components + Louvain community detection via Neo4j Graph Data Science to surface hidden fraud rings purely from shared-infrastructure graph topology — 100% recall on planted fraud clusters. Added a 3-layer GCN with focal loss for node-level fraud classification (AUC ~0.97 on imbalanced test sets).

Neo4j PyG GCN Focal Loss Community Detection Fraud Detection

Neo4j · Reinforcement Learning · Python · TACC HPC

Production-deployed knowledge graph modeling thousands of space objects and operators. Designed an RL-based behavioral pattern detector to autonomously identify regulatory non-compliance among adversarial actors operating with weak supervision — directly analogous to fraud ring detection in financial systems. Deployed on TACC Lonestar6 HPC, processing millions of satellite state records.

Neo4j Reinforcement Learning HPC Heterogeneous Graph Anomaly Detection

LangChain · Claude API · FAISS · Python

RAG-based Q&A system over structured document corpora: chunk-level retrieval, semantic reranking, and grounded answer generation with hallucination mitigation via source attribution and confidence thresholding. Designed for deployment in regulated legal/financial domains.

LangChain RAG Claude API FAISS Vector Search LLM

PyTorch Geometric · PyTorch · Python

Extracts 10 structural graph features per node from Neo4j (PageRank, Louvain community ID, shared-neighbor counts, WCC component size) and trains a 3-layer GCN with Focal Loss for fraud node classification under severe class imbalance. Full training loop with early stopping, cosine LR scheduling, and confusion matrix evaluation.

PyTorch Geometric GCN Focal Loss Graph Features Class Imbalance

Languages

ML / DL / Graph

Graph Databases & Tools

LLM / Agentic AI

Infra / DevOps

• 🎓 M.Sc, Aerospace Engineering — University of Texas at Austin (2017–2019)
• 🏛️ Graduate Archer Fellow, UT Austin (Summer 2026)
• 🌟 Brumley Next Generation Graduate Fellow, UT Austin (2024–25)
• 🥇 Gold Medalist — Best All-Rounder, NIT Rourkela (2013–17)
• 🇫🇷 Charpak Research Scholarship, French Government (2016)
• 🚀 Kalpana Chawla Award, SAE India Foundation (2016–17)