Sneha Maurya

• Developed a curated dataset of semantic code edits by mining and normalizing multi-file Java commits, using tree-sitter to extract AST structure and represent changes consistently across repositories. • Fine-tuned Qwen-3 Code(30B) with graph-aware attention and cross-file context encoding to improve next-action prediction in code completion. Developed a full evaluation pipeline using AST edit-distance, semantic checks, and compile-time validation to assess model behavior on large, real-world edit corpora. Capstone Project @ Data Science Institute | Columbia University Advisor - Atin Sood

As a TA I ran weekly lab sessions and open office hours where I walked students step-by-step through data manipulation in pandas, SQL joins, visualization, and building simple ML models for statistical problems. I reviewed code live in Jupyter notebooks, helped debug pipelines, and explained statistical ideas through hands-on examples so concepts clicked. I also prepared reusable Jupyter lab templates and short how-to guides for common issues, which made project iteration faster and reduced students’ debugging time.

I led two projects focused on improving tool reliability analysis and knowledge retention across multiple fabrication lines. I built a GenAI-assisted NLP workflow in Dataiku to cluster 10K+ EMS logs from CMP and Films tools, highlight recurring downtime themes, and automate failure summarization using prompt-engineered LLMs. I paired this with interactive Power BI Pareto dashboards to help engineers prioritize tool issues in real time. I also implemented an R Shiny ingestion pipeline to standardize 200+ Kaizen, 8D, and Lessons Learned reports into structured JSON and centralize them in Teradata, improving cross-line knowledge sharing and audit traceability.

Teaching Assistant II for two courses to MBAs and Executive MBAs at the department of Decision Risk and Operations. Both the courses are Taught by Professor Mattan Griffel. - Python for MBAs - Intro to Databases for Business Analysis

Led the development of a multimodal Retrieval-Augmented Generation (RAG) system to extract workforce-related insights from CSRD sustainability reports. Fine-tuned vision-language models including BGE-M3, Qwen2.5-VL, and GMM-based retrieval to process 300+ pages of complex PDF layouts—spanning OCR, tables, and diagrams with significant structured extraction accuracy. The system adapts to diverse reporting styles and regulatory queries, ensuring alignment with disclosure standards. Also built a Streamlit-based interface for query resolution and relevance visualization, significantly reducing manual review time.

• Built a GPT-4 based clinical NLP system that reads raw medical reports, retrieves relevant context with RAG, and generates clear, patient-friendly summaries; worked iteratively with clinicians to refine prompts and automated checks so outputs are clinically sensible and easy for patients to act on. • Engineered a secure, production AWS pipeline (S3 - Lambda - inference - versioned outputs) with CI/CD, monitoring and secrets/key management, and partnered with marketing to surface insights via SQL/Tableau dashboards for operational use.

Researched graph-based methods to sequence educational content by modeling prerequisite relationships and built a dataset with chapter subtitles to improve content organization.

Researched joint particle-picking and 3D orientation estimation for cryo-ET subtomograms under the guidance of Dr. Min Xu and Mostofa Rafid Uddin. Worked with GUM-net and downstream alignment methods, experimenting with rotation-equivariant feature learning and pose-refinement strategies to improve the robustness of 3D structural analysis.

Developed a YOLOv8-based kitchen safety monitoring system with fairness-aware detection and deployed real-time ML pipelines with reduced latency.

• Led the setup and optimization of Ambari clusters at Amazon Web Services, enhancing data processing performance and reliability. • Designed and deployed robust data pipelines, streamlining data flow for critical business applications. • Integrated Big Data tools like Hadoop, Kafka, and Hive, gaining hands-on expertise in scalable data management.

Fall (2025) Capstone (IBM Research) - Next action prediction in code generation High performing Machine Learning LLMs for Business Applications Spring(2025) Deep Learning for NLP Machine Learning for Data Science Statistical Inference and Modelling Advanced Spoken Language Processing Financial Structuring for Data Science Fall (2024) Applied Machine Learning Probability & Stat for Data Science Algorithms for Data Science Exploratory Data Analysis & Visualisations

Attended an intensive NUS lecture series (faculty + industry speakers) on deep learning, model efficiency, and deployment; completed hands-on AWS workshops covering data storage and inference pipelines. Built and presented a short project: trained a lightweight spatio-temporal model for webcam gesture recognition and demonstrated an end-to-end AWS pipeline (S3 → preprocessing → inference API) during the program.

Relevant Courses: Operating Systems, System Design, Database Management System, Data Structures and Algorithms, Statistical Machine Learning, Natural Language Processing, Software Management