V
Veditha Yechuri
About
Detail
Gujarat, India
Timeline
work
Job
school
Education
Résumé
Jobs
verified_user
0% verified
Machine Learning EngineerCitiusTechJan 2025 - Current (1 year 2 months)Developed ReX, a career-coaching GenAI app powered by GPT-4, Gemini, and LangChain, integrating RAG, Pinecone, and LlamaIndex boosted personalized response accuracy by 35% and drove a 30% increase in active users. Engineered scalable vector search pipelines using FAISS, ChromaDB, and OpenAI embeddings, enabling sub-second document retrieval across 50K+ enterprise knowledge base entries. Built CI/CD workflows using Docker, MLflow, and GitHub Actions, reducing deployment time from hours to minutes and improving rollback coverage by 60%. Containerized multi-LLM microservices via FastAPI and Kubernetes supported autoscaling inference endpoints across Azure Kubernetes Service (AKS). Deployed an end-to-end monitoring system using Prometheus and G- HMachine Learning EngineerHexaware Technologies, IndiaDec 2021 - Jul 2023 (1 year 8 months)Engineered unified credit-risk datasets using MySQL, Python, and Kafka, consolidating transactional and bureau streams, improving data completeness by 40% and enabling reliable modeling across lending systems. Developed advanced feature pipelines capturing repayment delays, spending patterns, and risk signals using Pandas and Scikit-learn, boosting model accuracy by 28% and enhancing default-prediction reliability. Built XGBoost, LightGBM, and CatBoost ensembles with SMOTE and temporal validation, reducing false negatives by 32% and strengthening early detection of fraudulent borrower behavior. Deployed real-time scoring microservices on Docker and AKS via FastAPI, delivering sub-200ms fraud alerts and enabling seamless integration with ent
- EMachine Learning EngineerEncode Testers, IndiaJun 2020 - Nov 2021 (1 year 6 months)Engineered and operationalized CNN-powered visual defect detectors using TensorFlow + TFLite on edge devices, improving inspection throughput by 40% in industrial IoT pipelines. Formulated an anomaly detection engine using Isolation Forests, Autoencoders, and Z-score monitoring to detect fraud in real-time, reducing false positives by 33%. Created a predictive maintenance framework for manufacturing equipment using LSTM-based time-series forecasting on Azure Databricks, improving downtime forecasts by 28%. Automated ETL workflows using Airflow and DVC, reducing data pipeline latency by 45% and enabling reproducible model training across diverse datasets. Optimized storage + retrieval in MongoDB and PostgreSQL using vector-based similarity s
Education
verified_user
0% verified
- UMasters in Data ScienceUniversity of Maryland Baltimore County, MDAug 2023 - May 2025 (1 year 10 months)
Projects (professional or personal)
verified_user
0% verified
- LLLM-based Blog SummarizationInstituted a multi-model blog summarization pipeline leveraging T5, BART, and Pegasus from Hugging Face Transformers, using ROUGE-L and BERTScore for performance ranking across tech, health, and education genres. Achieved a 20% reduction in reading time while preserving content coverage; Launched an interactive demo using Gradio and tracked model performance with Weights & Biases.
- NNLP-Based YouTube Comment ModerationProcessed 1M+ YouTube comments using spaCy, NLTK, and custom regex; Crafted a BERT-based toxicity classifier with a 94% F1-score. Implemented real-time moderation with FastAPI and Kafka, integrating with a live review dashboard.
- EEco-Scan Plant Disease DetectionDeveloped a mobile-ready plant disease diagnosis system using CNNs, FastAI, FastAPI, and Docker, integrated with MongoDB and Power BI dashboards, achieving 93% accuracy across 12+ classes and delivering the MVP 20% ahead of schedule with Agile (JIRA).