• Built CostCompass AI, a generative AI platform for infrastructure cost optimization that analyzed AWS usage and pricing data to deliver real-time anomaly detection, forecasting, and optimizations across AWS accounts and services

• Pioneered MCP + agent integration patterns, solving a fundamental deployment challenge adopted across Amazon AI initiatives; became the point of contact for MCP deployments across teams.

• Released the tool Amazon-wide, collaborating with an L8 principal engineer and presenting research org-wide; positioned to scale into a full service for executives, engineers, and financial analysts.

• Enabled success of LoadLens, an AI-powered load test report generator presented to CEO Andy Jassy, by applying the MCP + agent design pattern to unblock development and ensure production functionality.

• Overcame AWS Lambda runtime constraints (cold starts, dependency packaging, subprocess limits) by embedding MCP servers directly within Lambda and bridging them with environment variables, establishing a new serverless deployment pattern.

• Engineered enterprise-ready infra: auto-scaling ECS Fargate clusters, authentication via CloudFront + Lambda@Edge, multi-account IAM role assumption, and observability with CloudWatch.

• Developed a retrieval-augmented chatbot surfacing warehouse-specific information, improving customer engagement and lead conversion for Warehouse Exchange.

• Integrated with HubSpot CRM to optimize contact capture and retention workflows.

• Built FastAPI + Firebase backend for real-time chatbot handling, reducing latency and improving reliability.

• Implemented an analytics dashboard to track chat frequency, query types, and engagement peaks, giving sales teams actionable insights.

• Shipped a React/Next.js chatbot UI optimized for mobile/desktop and a mock chatbot tool for internal behavior testing.

• Engineered NLP + graph ML pipelines to build a physician–scientist knowledge graph from

OpenAlex/Dimensions.ai, stored in CockroachDB.

• Implemented an unsupervised author disambiguation system inspired by IEEE Big Data 2019 (HGCN embedding): constructed heterogeneous networks with CoAuthor/CoTitle/CoVenue relations, meta-path random walks, and graph-enhanced clustering.

• Built custom PyTorch Geometric (GCNConv) models over multi-relation graphs; trained with a skip-gram contrastive objective using alias sampling for positives/negatives.

• Integrated text embeddings (Word2Vec/Doc2Vec, NLTK preprocessing) with structural features, improving representation quality.

• Validated disambiguation with HAC, Louvain, KMeans clustering and PCA/t-SNE visualization; delivered an incremental update pipeline for streaming publications.

● Worked closely with senior members of DGV’s trading & research team to enhance existing trading strategies and improve investment outcomes for the firm’s investor base with assets in excess of $1.6 billion

● Engineered portfolio construction software in MATLAB, designed for extensibility and ease of modification, allowing users to input a set of security tickers and automatically generate optimized portfolios utilizing statistical optimization techniques

● Applied LSTM deep learning networks to forecast risk-adjusted returns to construct portfolios that often outperformed more traditional Markowitz based portfolio selection criteria

● Designed and back-tested multiple investment strategies based on the construction of optimal portfolios over a 15-year period; notably some strategies outperformed the S&P 500 Total Return Index in metrics such as Sharpe ratio and annualized risk vs returns

● Developed software using Python and Pandas to process historical options data from the Chicago Board of Options Exchange, reducing the dataset from 2.7 billion to 1 million relevant data points for further analysis while also optimizing for speed and accuracy

● Presented software architecture, research findings, and portfolio strategies to trading, research, and management departments, facilitating data-driven decision-making

● Developed smart contracts using technologies such as Solidity, OpenZepplin, Hardhat, and Curve.fi interface on the Ethereum blockchain

● Ensured correct smart contract behavior by writing test suites with Ethers.js and Mocha.js libraries

● Optimized the exchange rates of liquidity pools through mathematical analysis and simulation testing

● Developed simulation suites to simulate exchanges on Curve.fi decentralized exchange