Research interests: quantum machine learning and hybrid quantum–classical algorithms; multi-agent AI systems; applied machine learning in medical imaging and finance.
Education
New York University — MS, Computer Engineering
2023 – 2025
GPA 3.80 / 4.00. Coursework: Quantum Computing, Machine Learning, Deep Learning, Computer System Architecture, Network Security, Mathematics, Physics, Finance.
The National Institute of Engineering, Mysuru — BE, Computer Science
Research Intern, Center for Quantum Information Physics, NYU
JAN – MAY 2025
Advised by Prof. Javad Shabani.
Built a hybrid quantum–classical neural network estimating six-year lung-cancer risk from a single low-dose CT scan (3D volume), using Qiskit and PyTorch.
Prioritized faster inference at acceptable predictive accuracy, rather than outperforming existing models such as Sybil.
Papers & Technical Writing
Coin-Flip Pricing in a Tournament-Winner Market: A Paired-Contract Trade on the 2026 IPL Champion
2026
Independent writing. Bayesian analysis of a mispriced tournament-winner market: coin-flip pricing (0.75 combined) against a 93.3% base rate for a top-two seed, exploited via a paired position returning ~27.7%.
TicTacPro: Exactly Solving a Combinatorial Piece-Placement Game and Stress-Testing Deep Q-Learning Against It
2026
Independent writing. An exhaustive alpha-beta solver proves TicTacPro is a first-player win from all 27 openings — refuting the opposite conclusion reached by time-limited minimax self-play — and a Dueling Double DQN (792K parameters) trained with a curriculum of rule-based opponents reaches 100% against random and bullseye opponents, with pipeline pitfalls reported as negative results.
TicTacPro RL — a Dueling Double DQN trained with a curriculum of rule-based opponents, paired with an exact alpha-beta solve of the game. PyTorch, DQN, Alpha-Beta.
Dueling Double DQN (792K parameters) trained over 1M+ episodes on an NVIDIA GB10 DGX Spark (BF16, torch.compile, GPU-side prioritized replay, 8-fold symmetry augmentation).
An exhaustive alpha-beta solve proves a first-player win from all 27 openings; a curriculum of rule-based opponents lifted the DQN from 0/20 to 40/40 against the bullseye-fork strategy.
Multi-Step Tool Attacks — a Kaggle red-teaming entry against tool-using LLM agents (gpt-oss, gemma). Python, llama.cpp, LLM red-teaming.
Fires the exfiltration predicate with a benign http.post to a unique host; the guardrail inspects only the URL, so the untainted call passes while each post scores its own cell.
A tool-call prompt format and a forged-analysis multipost lifted throughput to a 73.88 public score; the hidden private guardrail blocked the line (private 0), a clean case of overfitting the visible guardrail.
Hybrid Quantum–Classical Neural Network — six-year lung-cancer risk estimation from a single low-dose CT scan. Qiskit, PyTorch.
Verified feasibility by over-fitting on a single image, given the exponential cost of operations on entangled qubits.
Used Qiskit to store model weights in qubits, later used for inference.
Stock Direction Prediction — fine-tuned FinBERT and LLMs feeding a Random Forest to forecast GameStop’s price direction. FinBERT, Llama, Random Forest.
Fine-tuned FinBERT, BERT, Gemma-7B, and Llama-11B to extract sentiment from Reddit data.
Sentiment and other features fed a Random Forest; price data from Yahoo Finance. Accuracy 0.59.
QubitQuery — retrieval-augmented assistant over lecture notes, built to make office hours more productive. LangChain, FAISS, OpenAI.
LangChain splits lecture notes into chunks, embedded via the OpenAI API and stored in a FAISS vector database.
Similarity search retrieves the most relevant chunks, which feed an LLM through a terminal-based interface.
chromeControl — MCP-compliant Chrome sidebar that answers questions about the current page. TypeScript, React, MCP.
Privacy-first safeguards: local processing, sensitive-data filtering, and domain-based security tiers.
Squaris — procedurally generated polycube packing puzzles over a single generation kernel (2D tiling, 3D lattice, Tetris-style). React, TypeScript, Three.js.
Reverse-construction generator guarantees solvability, with date-seeded daily puzzles.
Cube rotation group generated as the closure of three 90° axis rotations — 24 orientations per polycube, verified via BFS.
Jest suite covering generator invariants and rotation-group completeness; deployed on Vercel.
Industry Experience
Founding Engineer, Starboard
OCT 2025 – PRESENT
Applied research in multi-agent systems: agents with tool use that autonomously process shipment tracking and reconcile data across logistics platforms; fine-tuned open-source LLMs for document parsing.
Designed web scrapers and data pipelines extracting structured shipment data from carrier portals — 95%+ accuracy on real-time tracking, plus automated invoice reconciliation.
Fine-tuned open-source LLMs for entity extraction and document parsing, cutting average processing time from 2.5 minutes to under 10 seconds per document.
Software Development Intern, Tiny Archives
SEPT – DEC 2024
Backend archival systems for organizations and individuals, with a focus on scalability (Python, Django, PostgreSQL).
Software Development Engineer I, Finflux
NOV 2022 – JULY 2023
Built ELMS, an exposure and limit-management engine covering 35M+ customers.
Collaborated with backend engineers and joined client meetings to pin down and implement requirements.
Addressed Level 3 bugs, resolving production issues in the core Loan Management System (LMS).
Hackathons & Competitions
AI Agent Security — Multi-Step Tool Attacks — a Kaggle red-teaming competition; searched tool-using LLM agents for exfiltration chains, reaching a 73.88 public score.
HackNYU — an Acorns alternative with far more control over risk appetite and investment types.
Kiro Hackathon — chromeControl, an MCP-compliant Chrome AI assistant (see Selected Projects).
Reddit Devvit — Squaris, a spatial packing puzzle game (see Selected Projects).
YQuantum — attempted the BlueQubit problem of Peaked Circuits.
IMC Prosperity 3 — financial modeling for maximum profit across algorithmic and manual trading tracks.
Hackfinity — a RAG/LangChain chatbot helping students grasp hard-to-find topics.
Activities
Represented NYU in contract bridge (North American Bridge Championships, Philadelphia, 2025) and in poker at the Intercollegiate Poker Association annual tournament, 2025. Member, SASE and IEEE.