Sai Navya Jyesta

• Developed LLM-based pipelines using Python and FastAPI for processing vendor receipts and site analysis, extracting structured data from unstructured inputs and enabling automated approval workflows. • Built an application migration system using AST extraction with Neo4j and Milvus to analyze multi-language codebases and transform them into Java, capturing semantic dependencies across services. • Designed hybrid retrieval pipelines combining graph-based querying (Neo4j) and vector search (Milvus) to support context-aware code understanding and improve dependency resolution.•

• Designed a multimodal RAG system using Paligemma Vision and ColBERT late interaction to generate image embeddings for 1,000+ textbooks, achieving 170x speedup in semantic search over BERT-based models. • Utilized Qdrant multi-vector store for high-speed retrieval, reducing query time to sub-100ms. Incorporated LLaMA 3.2 Vision model for context-rich responses, enabling 128K token cross-chapter exploration. • Developed and deployed an MCP (Model Context Protocol) server to enable structured querying of a research database, reducing complex data access latency by 65% and improving interpretability and traceability of LLM-driven query workflows.

• In my role as a Research Assistant at Indiana University Bloomington in the Bashey Lab, I focused extensively on analyzing large-scale genomic data from 40 X. kop genomes. Utilizing advanced data analysis and bioinformatics tools like Anvio, I set up and managed various databases, including KEGG and NCBI COGs. • My work involved critical tasks such as Open Reading Frame (ORF) prediction to identify potential gene-coding regions, Hidden Markov Model (HMM) profiling to detect conserved protein sequences, and conducting phylogenomic analysis to explore evolutionary relationships within the data. I worked on generating graphs to compare genome assembly qualities and sizes using tools like Quast. • I developed and fine-tuned data analysis pipelines using tools like Anvio and Conda environments, ensuring the accuracy and reproducibility of our research results. This involved troubleshooting complex software issues and managing dependencies to enhance the reliability and efficiency of our data processing workflows. • I also utilized IU's Jetstream2, a cloud computing platform, to handle resource-intensive analyses, such as running pangenome workflows and generating ANI heatmaps. The optimizations I implemented contributed to a more streamlined analysis process, allowing our team to handle large datasets more effectively and efficiently. • I documented procedures and results to ensure that our methodologies could be easily reproduced and shared across the team. Through this role, I have honed my technical skills in Linux, Python, Docker, and various data analysis tools, which have been critical in conducting large-scale data projects. This experience has further strengthened my ability to manage and execute complex data analysis projects, directly contributing to the Bashey Lab's mission to advance our understanding of microbial diversity and evolution.

• Developed full-stack applications, including a Pantry Tracker and AI support platform using React, Next.js, and Firebase, improving task efficiency by 35%. • Engineered a SaaS flashcard system powered by OpenAI with real-time AI assistance, secure authentication, and integrated payment workflows, enhancing user engagement by 50%. • Advanced expertise in algorithms, data structures, and cloud technologies through hands-on projects and high-intensity hackathons, achieving top placements in multiple events.

• Developed scalable data pipelines using Azure Data Factory and Java, improving cross-platform query performance and system reliability for 10K+ daily records. • Contributed to the backend of a video streaming service, enhancing API performance and stability to support real-time content delivery for 1K+ concurrent users. • Automated real-time data sync between MongoDB and BigQuery using Azure Functions, reducing latency by 30% and enhancing backend efficiency. • Streamlined deployments with GitHub Actions-based CI/CD workflows, increasing release frequency by 30% and improving code quality and availability.

• Engineered financial risk detection pipelines using random forests and boosting, raising prediction accuracy from 79% to 93%. • Designed and deployed Spring Boot microservices and REST APIs to integrate fraud detection into enterprise ERP-like financial systems, ensuring high scalability and compliance with AML standards. • Enhanced system reliability through structured logging and real-time telemetry dashboards, participating in live site monitoring and incident resolution to ensure SLA compliance. • Refactored legacy Java code base into multi-threaded architectures, increasing transaction processing throughput by 35%.

• Processed and analyzed 75K+ daily transactions using Spark and Hive to identify behavioral patterns and simulate financial decision outcomes. • Applied panel data and instrumental variable models to evaluate pricing strategies, improving customer targeting precision by 15%. • Validated model performance with precision-recall metrics, reducing false positives in high-risk transaction classifications.