Renu Iswarya

AI Engineer

Tech Mahindra

Sep 2024 - Present

LLM Fine-Tuning for Domain-Specific Q&A System | Client: Microsoft — LLM + Fine-Tuning • Fine-tuned a pre-trained Large Language Model (LLM) on domain-specific datasets to improve response quality and domain adaptation using supervised learning techniques. • Built data preprocessing pipelines involving dataset cleaning, validation, transformation, and feature preparation to improve training data quality and model performance. • Implemented parameter-efficient fine-tuning techniques including LoRA and PEFT, optimizing computational efficiency while maintaining model accuracy. • Evaluated model performance using quantitative metrics and human evaluation, ensuring data quality, model reliability, and production readiness. • Deployed the trained model as a REST API for real-time inference, enabling scalable integration with enterprise applications and cloud-based services. Physics-Informed Neural Network for Hydraulic System Performance Prediction | Client: Parker • Developed a Physics-Informed Neural Network (PINN) for predicting pressure, flow, and temperature behavior in industrial hydraulic systems by integrating governing physical equations with deep learning models. • Performed data preprocessing, feature engineering, and sensor data integration while incorporating fluid dynamics constraints into the training objective to improve prediction accuracy with limited labeled data. • Implemented custom loss functions combining physics residual loss and data loss using automatic differentiation, optimizing model convergence and numerical stability during training. • Validated model performance against simulation and historical operational datasets using RMSE and MAE metrics, enabling predictive maintenance and improving equipment reliability for manufacturing operations. • Collaborated with cross-functional engineering teams to optimize model deployment workflows and support AI-driven decision making for industrial asset monitoring. Internal AI-Powered Project Management Chat Application | Internal Project — MCP + RAG • Developed an enterprise AI-powered knowledge platform using Model Context Protocol (MCP) and Retrieval-Augmented Generation (RAG), enabling scalable information retrieval across organizational knowledge sources. • Designed and implemented document ingestion, preprocessing, and indexing pipelines for structured and unstructured data using Python, OpenAI embeddings, and Pinecone to improve search accuracy and data accessibility. • Integrated Google Calendar API, Google Drive, JIRA, and external knowledge sources to automate enterprise workflows and streamline data processing across multiple systems. • Developed REST-based backend services for data retrieval, document processing, and real-time query execution while following scalable software engineering practices.