Kaushal Rathi

AI Research Engineer at Infosys

Led end-to-end design and implementation of a compliance verification multi-agent system for the EU AI Act using LangGraph, Python, and RAG, integrating with OpenAI ChatGPT, Claude, and Gemini LLMs; achieved 96% backlog resolution (1440/1500) and reduced verification time per project by ~88%. Directed a cross-functional team to develop an automation pipeline for the SDLC of web-based applications via a multi-agent system that generated requirements, user stories, and architecture designs, integrated with Jira, and deployed code to local or GitHub environments, cutting development time by ~90%. Led Apple’s AI policies refinement and red-teaming efforts across applications (text + image); built CI/CD automated red-teaming pipelines using Garak, PyRIT, and Python; identified 6 major policy gaps and 1,000+ safety issues. Implemented text data annotation frameworks and embeddings-based similarity matching (KNN, fine-tuning SLMs like Gemma, Mistral) to categorize ~62K prompts under company

Machine Learning Engineer at ALLTRNA

Developed a robust ML/DL regression pipeline (BNN, CNN, ElasticNet, RandomForest, XGBoost, ConvLSTM1D, TCN+CNN, Transformers+NN) using TensorFlow, Keras, and Scikit-learn; accelerated in vitro research by up to 50% and saved an estimated $450K in downstream lab costs via earlier identification of viable drug candidates. Established scalable MLOps architecture (MLFlow, DVC and stress tests), enabling CI/CD for regression models and reducing deployment cycles by 35%. Implemented advanced hyperparameter tuning (nested cross-validation, GridSearchCV, GASearchCV) and evaluated with metrics (F1, confusion matrices) to increase model accuracy by ~20%. Investigated epistemic and aleatoric uncertainty in Bayesian neural network regression to enhance interpretability. Performed comprehensive data processing (EDA, ETL) and built configurable Pandas-based pipelines, reducing manual data prep time by ~40%.

Machine Learning Engineer at Precision Technologies Corporation

Owned end-to-end development from architectural streamlining to cloud deployment; leveraged Scikit-learn (decision trees, random forests, XGBoost, Bayes classifiers) to solve a 650K vendor-options mapping problem. Orchestrated data pipelines (ETL, data wrangling, statistical analysis, visualization) and feature engineering with Pandas; incorporated MLOps features such as monitoring, auto data collection, model training, testing, and deployment.

Software, Computing, and AI Supplemental Intern at Ford Motor Company

Researched and architected synchronous and asynchronous federated learning solutions for autonomous driving; prototyped FedAvg and FedBuff in Python (TensorFlow, Keras) on SK-TDA4VM boards. Presented and defended findings to managers at various levels, earning a perfect score (6/6) and positive stakeholder feedback. Received the Ford 2022 Signature Event top achiever badge.

Machine Learning Intern at Cyronics Instruments

Detected ECG abnormalities (myocardial infarction, T-wave elevations) using LSTM and 1D CNN; explored HMM and class activation maps. Built time-series forecasting with LSTM reducing oil engine testing cycle time by ~83%. Deployed object detection/instance segmentation (Mask R-CNN) for fault detection on moving conveyors.