I am developing END-to-END GenAI/LLM/ML products, including RAG, question answering, conversational AI (Chatbots), text summarization, named entity recognition (NER), text classification, sentimentanalysis, and machine translation. Projects:1st Project: Created an End‑to‑End Advanced Retrieval Augmented Generation (RAG) Chatbotbased on query expansion (prompt engineering), ensemble retrieval, and reranker. Results: Reduced hallucinations from 20% to 5%, and latency from 100s to 15s. Tools: LangChain, Unstructured, ChromaDB/FAISS, HuggingFace, Docker, AWS EC2, Git, and Streamlit/FLASK.• 2nd Project: Created an End‑to‑End LLM web‑app for retail searching products via fine‑tuning GPT‑3.5 Turbo and coupling it with SQL database using Langchain. I used FastAPI and Jinja for web‑interface, and deployed the web‑app on Azure with CI/CD on GitHub actions.• 3rd Project: Developed an End‑to‑End Generative AI web‑app based on google‑flan‑t5‑base LLM and prompt engineering to answer questions given document contexts. I utilized the following packages: Langchain , FAISS, HuggingFace, Docker, AWS EC2, GitHub action, and Streamlit.• 4th Project: Developed an End‑to‑End LLM web‑app via LaMini‑Flan‑T5‑248M to perform text summarization of PDF contexts, using HuggingFace, Langchain, Docker, AWS EC2, Github action (CI/CD), and Flask.• 5th Project : Pretrained a sparse transformer variant for sequence modeling based on local attention. Results: A remarkable 78% reduction in the FLOPs computation time compared to the vanilla attention. The model gives 2.8 Perplexity (PPL) in evaluation (on text_8 benchmark), and generates a novel coherent text for up to 100 tokens.

Roles: 1. Gathering, cleaning, and organizing datasets.2. Identifying and addressing data quality issues and inconsistencies.3. Developing and training predictive models using supervised and unsupervised machine learning algorithms.4. Applying statistical methods to derive meaningful insights and validate hypotheses.5. Monitoring model performance over time and making necessary adjustments.Projects:1. 1st project: modeling and analyzing the performance of an active flow control system using deep neural networks (DNNs)2. 2nd project: machine learning for optimal flow control in axial compressor via deep neural networks (DNNs) and genetic algorithms (GAs)3. 3rd project: developing a physics-informed neural networks (PINNs) model to predict andanalyze the performance of a flow field inside a centrifugal pump.

Roles:1. Applying statistical methods to derive meaningful insights and validate hypotheses.2. Conducting hypothesis testing and interpreting results.3. Identifying relevant features and reducing dimensionality to improve model efficiency via principal component analysis (PCA) and dynamic mode decomposition (DMD).3. Developing and training predictive models.Project: 1. Predicting and analyzing the performance of a thermoacoustic refrigerator via response surface methodology (RSM) and deep neural networks (DNNs)

Roles:1. Developed a deep reinforcement learning (DRL) algorithm integrated with a computational fluid dynamics environment called FEnics, to explore active flow control strategies around a circular cylinder using unsteady-mode plasma actuators.2. Conducted independent research studies involving artificial neural networks (ANNs), convolutional neuralnetworks (CNNs), and recurrent neural networks (RNNs) in the areas of computer vision and natural language processing.Project:1. Deep Reinforcement Learning for Active Flow Control around a Circular Cylinder Using Un-steady-mode Plasma Actuators.