Harshith Kori Raj

• Designed and implemented LLM-powered AI agents using LangChain/LangGraph for multi-step reasoning and automation. • Built RAG pipelines integrating vector stores, prompt orchestration, and tool-calling for domain workflows. • Developed APIs for agent services with FastAPI, enabling scalable inference and conversation memory. • Optimized prompts, evaluation metrics, and guardrails to improve reliability and reduce hallucinations.

• Developed ML models for predictive analysis using Python, Scikit-learn & TensorFlow, improving accuracy through tuning. • Built data preprocessing pipelines with Pandas/NumPy and evaluated models using cross-validation metrics.

Contributed to the development of three research papers, including one published in IEEE. Responsibilities included dataset curation, implementing DWT, DCT, SVD, GAN-based person re-identification, hybrid-domain steganography, statistical feature engineering, and designing ANN/CNN classifiers. Conducted experiments, performance evaluation (PSNR, SSIM, IoU, accuracy), visualization, and drafting of methodology and results sections for publication.

Co-authored research on Electronic Health Record (EHR) analytics, performing data cleaning, normalization, and feature engineering, and developing a CNN-based prediction model that achieved enhanced accuracy with early convergence and plateau stabilization.

• Built ML pipelines for data preprocessing, feature engineering, and model training using Python, Scikit-learn & TensorFlow. • Developed predictive models achieving improved accuracy through hyperparameter tuning and cross-validation. • Automated data workflows with Pandas & NumPy, reducing manual processing time significantly. • Collaborated on model evaluation, visualization, and deployment using Flask for inference APIs.

Built end-to-end web applications using React, Node.js, and MongoDB, improving feature delivery speed by 30%. • Designed RESTful APIs and authentication workflows with JWT & Express, enhancing system security. • Implemented responsive UI components and optimized performance using lazy loading and caching.