• Built an AI-powered document ingestion agent that automatically parses uploaded files and generates purchase orders within the Order Management System (OMS), reducing manual processing effort and improving order handling time by ~45%

• Designed and implemented Spring Boot microservices with RESTful APIs and Hibernate ORM, optimizing database indexing strategies and connection pooling to reduce API latency by 40% and improve service responsiveness

• Designed and implemented an LLM evaluation framework using LangSmith, OpenRAG-Eval, and Azure OpenAI to systematically benchmark model responses across 1,500+ queries per evaluation cycle, identifying grounding gaps, retrieval failures, and hallucination patterns

• Developed automated evaluation pipelines incorporating metrics such as UMBRELA, AutoNugget, Citation Support, and Precision/Recall to measure retrieval relevance, contextual alignment, and citation validity in RAG-based systems

• Built scalable data processing pipelines on GCP using BigQuery and Vertex AI, reducing end-to-end analysis time from ~5 hours to under 30 minutes and enabling rapid iteration on model evaluation experiments

• Accelerated backend engineering workflows using AI-assisted development tools (Cursor, Claude) for code generation, debugging, and refactoring, improving development velocity and reducing implementation time by ~60%.

• Built and maintained high-performance Golang backend APIs powering 20+ security and application monitoring dashboard widgets, designing custom Elasticsearch queries capable of analyzing 300GB+ telemetry datasets with sub-2 second response times.

• Modernized backend infrastructure by refactoring 40+ Go microservices and migrating APIs to gRPC, strengthening service-to-service communication, addressing security vulnerabilities, and ensuring FIPS 140 compliant builds using Bazel

• Crafted robust backend services for SRX firewall management, processing \textbf{5,000+} configuration changes daily with high throughput and reliability across hybrid networks

• Improved scalability of distributed services by implementing data partitioning and sharding strategies, enabling horizontal scaling across multiple nodes and supporting a 3× increase in telemetry ingestion without performance degradation

• Built a Natural Language → SQL querying system using Google Generative AI embeddings and a Chroma vector store, enabling non-technical users to retrieve database insights through plain-English queries with dynamic table selection and contextual query memory

• Developed an AI-powered research analysis tool using the OpenAI API to automatically scan academic papers, detect methodological issues, and surface potential weaknesses, improving peer-review acceptance rates by ~15% across conference submissions

• Implemented CI/CD automation with GitHub Actions and a real-time monitoring dashboard to analyze pull requests, flag potential issues, and provide visual insights for developers, enabling team-wide adoption and production-ready workflows

• Designed and trained a Named Entity Recognition (NER) pipeline to extract entities such as contracts, organizations, and key dates from user queries, significantly improving search accuracy and contextual relevance in downstream retrieval systems

• Engineered the Activity Manager module for Citrix Workspace, a centralized session monitoring system adopted by 3,000+ enterprise customers, contributing to $750K+ in revenue growth by improving observability and operational efficiency

• Developed a workspace scheduling feature that automates desktop lifecycle management, reducing unnecessary cloud VM usage and operational costs; the solution won 1st place at a company-wide hackathon

• Built and deployed production-grade Workspace UI modules using Next.js, Vue.js, Tailwind CSS, Node.js, MongoDB, and MySQL, enabling modular frontend architecture and seamless backend integration through event-driven services

• Improved frontend performance of Citrix Workspace by implementing Redux Toolkit, React Query, memoization, lazy loading, debouncing, and service workers, reducing component render time by 50% and improving perceived page load speed to ~0.5s (80% faster)