Keerthi

Sr. Data Scientist and AI/ML Engineer with 10 years of proven experience in designing and deploying end-to-end machine learning and deep learning solutions across healthcare, banking and enterprise domains. Adept in building scalable data pipelines, predictive models, and advanced analytics workflows involving NLP, computer vision, and time series forecasting.

Proficient in Python, SQL, and R with hands-on expertise in TensorFlow, PyTorch, Scikit-learn, and MLOps tools such as Docker, Kubernetes, MLflow, and DVC. In the past year, led projects in generative AI using LLMs (GPT-4, LLaMA 3, Gemini), RAG architectures, Hugging Face Transformers, and LangChain deployed on cloud platforms such as Microsoft Azure and AWS, with integrated vector search using FAISS and Pinecone. Skilled in delivering explainable, secure, and regulatory-compliant (HIPAA/GDPR) AI solutions.

Passionate about solving real-world problems through responsible AI and innovation at scale. And always eager to learn new technologies, adaptable, hardworking, and committed to continually upgrading skills to stay at the forefront of AI/ML advancements.

AI/ML ENGINEER - The Vanguard Group - TX

(2025-03)

Designed and deployed a scalable AI/ML platform for investment intelligence, predictive analytics, anomaly detection, and regulatory knowledge management, leveraging machine learning, deep learning, Generative AI, LLMs, and RAG. Built AWS-based ML pipelines, real-time anomaly detection capabilities, and an internal RAG-powered Q&A assistant to enable analysts to efficiently query regulatory documents and fund prospectuses. Implemented MLOps practices including model versioning, A/B testing, drift monitoring, and automated retraining to support reliable production AI systems.

• Designed and deployed machine learning models for predictive analytics, classification, and recommendation systems, improving business decision-making and operational efficiency.
• Developed and optimized deep learning models using TensorFlow, PyTorch, and Scikit-learn for real-world AI applications.
• Used SHAP and LIME to explain predictions from transformer and Gen AI models for anomaly detection and maintenance, helping stakeholders understand model decisions and build trust.
• Designed and deployed Generative AI (Gen AI) and LLM-powered solutions using Hugging Face Transformers and GPT-4 APIs to automate smart building operations, enhance energy efficiency insights, and enable intelligent fault diagnostics.
• Built end-to-end ML pipelines, including data collection, preprocessing, feature engineering, model training, evaluation, and deployment.
• Implemented Large Language Model (LLM) solutions using GPT, Claude, and open-source foundation models for content generation and automation.
• Fine-tuned transformer-based models using domain-specific datasets to improve response accuracy and task performance.
• Fine-tuned transformer models (BERT, RoBERTa) and domain-specific Gen AI models using PyTorch to support tasks such as anomaly detection, system state classification, and predictive maintenance, achieving a 22% improvement in predictive maintenance scheduling accuracy.
• Developed Retrieval-Augmented Generation (RAG) applications integrating vector databases such as Pinecone, FAISS, and ChromaDB.
• Created AI-powered chatbots and virtual assistants using NLP techniques, reducing manual effort and enhancing user experience.
• Designed real-time anomaly detection system on Kafka streams to flag suspicious trading activity, integrating with AWS SageMaker for auto-retraining triggered by data drift metrics.
• Led development of internal RAG-based Q&A assistant using LangChain + Claude API, enabling analysts to query regulatory documents and fund prospectuses in natural language; adopted by 200+ internal users within 3 months of launch.
• Collaborated with quant research team to build factor-based return forecasting models, contributing directly to improved fund performance benchmarking workflows.
• Established MLOps best practices: model versioning, A/B testing framework, and automated drift monitoring using MLflow and custom dashboards, reducing model degradation incidents.
• Mentored 2 junior ML engineers and led bi-weekly knowledge-sharing sessions on LLM fine-tuning and responsible AI practices.

Data Scientist with AI/ML ENGINEER - JOHNSON CONTROLS - TX

(2023-10 - 2025-02)

Designed and deployed a cloud-native AI platform at Johnson Controls to automate smart building operations, leveraging Gen AI, LLMs, RAG pipelines, and Azure cloud services. The solution enabled real-time fault diagnostics, predictive maintenance scheduling, and contextual decision support for field technicians. Integrated scalable MLOps pipelines, ensured GDPR and ISO 27001 compliance, and achieved measurable improvements in operational efficiency, energy optimization, and system uptime across enterprise building infrastructures.

• Developed advanced Python-based ML pipelines for training, evaluation, fine-tuning, and deployment of AI/ML models supporting large-scale enterprise workloads.
• Led sprint planning, technical roadmap alignment, and cross-functional stakeholder coordination to ensure timely delivery of ML initiatives.
• Mentored junior engineers and led technical design discussions to promote scalable ML architecture best practices.
• Provisioned scalable Azure ML and AKS environments using Infrastructure as Code principles, ensuring reproducible and version-controlled deployments.
• Implemented vector similarity search systems using vector databases to enable high-performance semantic retrieval and contextual knowledge discovery.
• Managed model artifacts and container images using structured versioning and repository-based distribution strategies to ensure reproducibility across environments.
• Applied advanced statistical modeling techniques including regression diagnostics, multicollinearity analysis, and probabilistic modeling to improve model interpretability and reliability.
• Implemented production monitoring for model drift, latency, and SLA adherence with automated alerts and rollback mechanisms to ensure system reliability.
• Owned the complete ML lifecycle from problem definition and data validation to deployment, monitoring, and business KPI tracking in production environments.
• Built and maintained Retrieval-Augmented Generation (RAG) pipelines using LangChain, FAISS, and Pinecone, improving contextual search precision by 45% across building manuals, sensor logs, and maintenance protocols.
• Optimized and exported models using TensorFlow Lite for edge deployment with reduced latency and memory footprint.
• Implemented distributed model training using multi-GPU configurations to accelerate training time and improve scalability.
• Developed production-grade Python microservices for ML inference with modular, maintainable, and deployment-ready code supporting high-throughput AI workloads.
• Implemented prompt engineering and prompt optimization strategies to improve LLM response quality, reduce hallucinations, and enhance task-specific performance.
• Designed and deployed Document AI pipelines integrating OCR, text extraction, document parsing, normalization, and structured data transformation for large-scale document processing.
• Applied post-training quantization and parameter-efficient fine-tuning (LoRA) to transformer models, reducing inference latency and optimizing memory utilization for production-grade deployments.
• Developed and integrated real-time vector similarity search systems using SentenceTransformers and OpenAI Embeddings, supporting fault resolution dashboards and technician decision-making.
• Deployed scalable Gen AI, ML, and LLM-based microservices using Azure Machine Learning, Azure Kubernetes Service (AKS), and Azure Functions, achieving 99.9% uptime and reducing model inference latency by 35%.
• Extended MLOps pipelines with MLflow, DVC, Docker, and Kubernetes to automate model versioning, drift monitoring, retraining, and CI/CD deployment, reducing manual retraining effort by 40%.
• Designed and deployed a cloud-native architecture on Azure, leveraging Azure ML and AKS to ensure high availability, seamless scalability, and built-in disaster recovery for critical smart building operations.
• Engineered prompt-optimized APIs and fine-tuned Generative AI and LLMs (GPT-3.5, LLaMA 3, Gemini) for AI-powered dashboards and voice/chat-enabled support systems used by field technicians and facility managers.