Headline

Senior GenAI/ML Engineer @ EY | LLM Fine-Tuning & Multi-Agent Systems | GenAI, RAG, LangGraph, PyTorch | Ex-Cerebrone.ai

About

I build production LLM and agentic AI systems from distributed fine-tuning pipelines to real-time voice agents currently as a Senior AI & ML Engineer at EY. Over the last 4+ years I've moved from data science fundamentals into deep GenAI/LLM engineering, shipping systems including: • A distributed LLM fine-tuning framework (PyTorch, QLoRA, multi-node GCP training) that cut infra costs 40–60% while training 7B+ parameter models on commodity GPU clusters • A real-time agentic voice assistant (FastAPI, Deepgram, Twilio, WebSockets) with an LLM-as-a-Judge evaluation pipeline for speech quality • A multi-agent healthcare assistant (LangGraph + RAG) that routes users across appointment booking, cancellation, and rescheduling workflows • A brand-compliance agent using RAG + vector search to auto-score marketing content against guidelines, deployed on Azure Serverless Before AI/ML, I earned an M.E. in Power & Energy Systems and co-authored two IEEE/journal publications on power-system optimization a background that still shapes how I approach ML problems: rigorously, with a bias toward measurable impact over model complexity. Core stack: Python, PyTorch, LangChain/LangGraph, RAG pipelines, Docker/Kubernetes, AWS/GCP/Azure, FastAPI, PostgreSQL. Open to conversations on LLM engineering, agentic AI systems, and applied GenAI in enterprise settings. Reach me via LinkedIn or Reach me at iqbalmohammedshuaib2@gmail.com

Location

Greater Bengaluru Area

AI stack

claude code

claude

codex

cursor

mcp

Glean

Work experience

Senior GenAI/ML Researcher and Engineer at EY, Feb 2025 - Present · 1 yr 8 mos, Bengaluru, Karnataka, India, • Cut LLM training infrastructure costs 40–60% by architecting a memory-efficient, multi-node distributed fine-tuning framework (Python, PyTorch, QLoRA, torchrun) on GCP, training 7B+ parameter models across 200K+ samples on commodity multi-GPU clusters. • Built and shipped a real-time agentic voice AI assistant (STT/TTS) using FastAPI, Deepgram, and Twilio over async WebSockets; added an LLM-as-a-Judge evaluation pipeline (LangSmith, Ollama, Scikit-learn) to benchmark speech quality and curate gold-standard fine-tuning datasets, containerized on Docker and deployed via AWS/Kubernetes. • Automated brand-compliance review at scale with an LLM-powered agent (LangGraph, RAG, vector DBs) that classifies and scores influencer/marketing content against brand guidelines, deployed serverlessly on Azure + FastAPI. • Designed a multi-agent healthcare assistant (LangGraph, FastAPI) with supervisor-based orchestration routing users to specialized agents for appointment booking, cancellation, and rescheduling, combining RAG retrieval and live web search to improve response accuracy.

AI & ML Developer at Cerebrone.ai, Jan 2022 - Feb 2025 · 3 yrs 2 mos, Remote, Generative AI, AI Agents, Langsmith, LangGraph, Autogen, CrewAI, Langchain, Fine Tuning LLMs such as Phi2, Llama2, Llama3.1 for custom datasets. • Developed an advanced Agentic Retrieval-Augmented Generation (RAG) system for analyzing customer issues across brands, leveraging CrewAI, GPT-based decision-making agents, and Python to autonomously process complex queries, generate diagnostic insights, and produce actionable outcomes, improving customer support analysis and engagement. • Build an AI-powered Medical Diagnostics Support Application leveraging LLMs, LangChain, and Flask, enabling real-time medical document processing and diagnosis assistance by integrating vector databases, PDF/Text ingestion, and cloud-based storage solutions, enhancing diagnostic accuracy and accessibility. • Designed and implemented a modular workflow using LangGraph to extract technical skills and generate ATS optimized resume points from job descriptions. Leveraged advanced AI models (GPT-4) for skill extraction and validation, and integrated Streamlit for an interactive user interface, enhancing recruitment efficiency through tailored, contextually relevant resume content generation. • Developed an AI-powered Medical Image Analysis application utilizing GPT-4 Vision, OpenAI Whisper, and Gradio to provide real-time diagnostic insights, treatment recommendations, and follow-up care plans, leveraging LangChain and ChromaDB for scalable data processing and storage, significantly enhancing accessibility and efficiency in healthcare diagnostics. • Streamlined operational capacity for generating tailored resumes using the Phi-3 model with targeted enhancements from LoRA and QLoRA applications, leading to a consistent reduction in average response time by 60%.

Data Scientist intern at Trainity, Aug 2021 - Dec 2021 · 5 mos, Remote, Developed a lightweight Multimodal AI application integrating image-to-text generation, speech transcription, and text-to-speech synthesis using Llava 7B, Whisper, and gTTS, with an interactive Gradio-based user interface, achieving optimized performance through 4-bit quantization for scalable deployment. Developed a RAG-Based Knowledge Management System leveraging LangChain, Vector Databases, and LLMs to enable intelligent document processing, automated knowledge retrieval, and contextual Q&A capabilities, optimizing enterprise information accessibility and decisionmaking efficiency. Designed an end-to-end machine learning solution for Credit card default prediction, achieving a high accuracy (82%) through advanced model optimization techniques, including hyperparameter tuning for XGBoost. Streamlined the pipeline with data pre-processing (scaling, feature selection), clustering, and automated model evaluation. Developed an end-to-end Waste detection system leveraging YOLOv3, enabling real-time object detection with optimized pipelines for data preprocessing, model training, and deployment through a scalable Flask application.

Data Science Intern at Internshala Trainings, Apr 2021 - Jul 2021 · 4 mos, India, My abilities as a data scientist are built, tested, and improved upon by the Data Science Specialization offered by Internshala Trainings. During this extensive program, I learned Advanced Excel, Python, SQL, Machine learning, and Tableau through lectures, live projects, regular vivas, and live classes. I spent my days working on extensive problems, e.g., Taxi scheduling & fare problems for a cab service firm, Predictive analytics on a dataset for FIFA19, and File handling and String Manipulation using Python. Did well in the test and Projects that are conducted by Internshala Trainings.

Data Science Intern at Internshala India, Apr 2021 - Jul 2021 · 4 mos, Optimized Student academic performance prediction by developing a pipeline to preprocess diverse datasets (one-hot encoding and standardization) and training advanced ML models like CatBoost, achieving up to 88.07% R² accuracy. Conducted rigorous evaluation and visualized results to enhance model interpretability, accuracy, yielding actionable insights into academic trends. Engineered aHandwritten Digit Recognition System with a CNN(Convolutional Neural Network) trained on theMNIST dataset, achieving 97.8% accuracy across 60,000 training and 10,000 test samples, demonstrating advanced machine learning expertise. Conducted comprehensive exploratory data analysis for Zomato and developed a predictive model for restaurant cost estimation. Achieved a remarkable R² accuracy of 99.99% using an Extra Trees Regressor, outperforming other models like Linear Regression, Decision Tree, and Random Forest through advanced feature engineering and hyperparameter optimization.

Education

Master of Engineering, Energy Management and Systems Technology/Technician, University Visvesvaraya College of Engineering ( UVCE )

Data Analytics, Trainity

Public links

github: https://github.com/Shuaib11-Github