Snehit Vaddi

Machine Learning and Data Engineer

2.5+ years of experience developing and deploying ML solutions for AT&T. Skilled in building ML pipelines, ETL with PySpark, SQL, and Azure cloud's da...

About

2.5+ years of experience developing and deploying ML solutions for AT&T. Skilled in building ML pipelines, ETL with PySpark, SQL, and Azure cloud's data engineering tech stack. Currently building a Multimodal RAG application for a large enterprise. Published researcher in deep learning.

Experience

Feb 2023 - Dec 2024

Graduate Research Assistant·University of Florida

  • Led the development of Citrus-Ecoli-FluoroNet, achieving 90% accuracy in E.coli classification
  • Implemented MLOps practices with MLflow and Docker for reproducibility
  • Engineered explainable AI techniques with YOLOv8 Eigen-CAM
  • Designed MongoDB schema for 100GB+ metadata, optimizing retrieval by 25%
  • Implemented A/B testing for ML models evaluation
May 2024 - July 2024

AI Software Developer Intern·GeoSpider

  • Developed GenAI-powered support system with FastAPI and LangChain
  • Implemented FAISS vector search with HNSW algorithm
  • Created real-time LLM API interface for improved user experience
Jun 2021 - Dec 2022

Data Engineer·AT&T (with Accenture)

  • Built NLP pipelines for technician dispatch prediction
  • Developed ML ensemble model saving $2M annually
  • Optimized ETL workflows with Azure and PySpark
  • Improved data processing time by 30% with PySpark optimization
  • Migrated to Elasticsearch for enhanced log analytics

Skills

Machine Learning & GenAI

TensorFlowKerasScikit-learnOpenCVNLTKHugging FaceLangChainOpenAI API

Cloud Platforms

Apache Spark (PySparkSQLSpark MLlib)Azure DatabricksAzure Data FactoryAWS (S3Athena)

Programming Languages and Tools

PythonSQLGitDockerPostmanJupyter NotebooksVS Code

Projects

RAG Anything

Developed a RAG document chat application, leveraging LangChain, SQLDatabaseChain, OpenAI embeddings, and FAISS, implementing control flow for question-based LLM selection.

LangChainRAGWord EmbeddingsVector Stores

Idiomatically-Llama

Fine-tuned LLama2 7B model for idiom enhancement, built RAG pipeline with LangChain and CromaDB, achieved 0.91 ROUGE-L F1 score using 4-bit quantization.

QuantizationOpenAIHugging FaceModel fine tuning

Education

Jan 2023 - Dec 2024

Master of Computer & Information Science

University of Florida

GPA: 3.5/4

May 2017 - Jun 2021

Bachelor of Technology in Computer Science

GITAM University

GPA: 3.8/4