Tejeswar Reddy Nalijeni

About Tejeswar Reddy Nalijeni

Experienced ML/AI Engineer with 3+ years of end-to-end experience designing, building, and deploying production-grade machine learning and AI solutions across insurance, banking, and technology domains. Expert in developing Retrieval-Augmented Generation (RAG) chatbots powered by LLMs, vector databases (Pinecone), and prompt-driven workflows to deliver context-aware customer support and reduce resolution times. Proven track record in credit-risk and churn-prediction modeling using XGBoost, Random Forest, ensemble methods, and advanced resampling techniques (SMOTE), achieving double-digit improvements in recall and AUC-ROC. Skilled in constructing robust ML pipelines Terraform for infrastructure as code, MLflow for experiment tracking, Jenkins/Git/Pytest for CI/CD to automate training, deployment, monitoring, and seamless rollback. Adept at feature engineering, data preprocessing (KNN imputation, IQR filtering, StandardScaler, one-hot encoding), and EDA with Pandas, NumPy, and Matplotlib to uncover insights and drive model performance. Experienced in fine-tuning GPT-based LLMs on proprietary corpora using supervised methods and LoRA adapters, incorporating user feedback via Pipelines to continuously refine conversational accuracy. Strong collaborator and communicator mentored students in Python and Big Data, published peer-reviewed research on imaging analytics, and translated technical findings into actionable strategies for stakeholder.

Experience

1. Ai Engineer

   Allstate

   May 2024 — Present · US

   Implemented an AI-driven policyholder-assistance chatbot using LangChain for Retrieval-Augmented Generation (RAG) with custom agent workflows, enabling context-aware, human-like support across call-center and digital channels, and reducing average resolution time for routine inquiries. Built a Python-based data ingestion pipeline to load customer interaction logs and policy documents into a secure cloud repository, applied text-cleaning and normalization routines, and split content into manageable chunks using a RecursiveCharacterTextSplitter before generating high-dimensional embeddings for fast, accurate semantic retrieval. Integrated Pinecone as a high-performance vector store, architected prompt-driven retrieval strategies to filter and rank the most relevant document segments, and orchestrated calls to GPT models for timely, accurate response generation. Fine-tuned GPT-based LLMs on proprietary insurance corpora including policy manuals, claim narratives, and underwriting guidelines using supervised fine-tuning and parameter-efficient LoRA adapters, enhancing domain-specific understanding, reducing hallucinations, and boosting response accuracy by 20%. Designed and executed end-to-end evaluation pipelines with Promptfoo to automate prompt benchmarking across real-world client inquiries, measure relevance and accuracy metrics (BLEU), and iteratively improve conversational quality. Developed a product-recommendation engine for insurance offerings by prototyping collaborative and content-based filtering techniques, scaling to TensorFlow-based deep-learning architectures, and driving measurable increases in cross-sell and upsell conversion rates. Constructed claims-frequency forecasting and dynamic-premium models using NumPy-driven analytics, conducting back-testing to validate predictive performance, and delivering data-driven insights for pricing strategies and capacity planning.

Education

• University of Cincinnati

  Master of Science - MS

  2023 — 2024

• Saveetha School of Engineering

  Bachelor of Engineering - BE

  2017 — 2021