Pushkara Chakka

Led the development of an enterprise-scale Generative AI-enabled Supply Chain Optimization Platform to modernize retail operations for over 2,200 stores. Designed advanced demand forecasting models leveraging Temporal Fusion Transformers, LSTMs, and Prophet, reducing forecasting errors by 25%. Built inventory optimization using Hugging Face Transformers and generative simulations, reducing surplus inventory by 17%. Developed logistics intelligence models combining geo-spatial ML with LLM-powered extraction improving delivery reliability and cutting last-mile costs by 14%. Enhanced pricing models with LightGBM, TensorFlow, and RAG pipelines, increasing promotional ROI by 20%. Engineered real-time data pipelines for structured and unstructured data ingestion and deployed ML and Generative AI services using modern MLOps practices including MLflow, Docker, and Azure Kubernetes Service. Delivered GenAI-powered BI dashboards and embedded assistants to accelerate executive decision-making and

Led end-to-end design, deployment, and optimization of AI/ML and Generative AI systems for statewide traffic safety operations. Improved incident detection accuracy by 25% using YOLOv8, Faster R-CNN, and DeepSORT on live CCTV streams. Built anomaly detection with LSTM Autoencoder and Isolation Forest on IoT sensor data providing real-time alerts under 10 seconds. Deployed streaming ML pipelines on Azure Gov Cloud processing over 1 million events/day with 99.9% uptime. Created edge AI inference pipelines on NVIDIA Jetson to reduce latency by 40%. Developed RAG-based generative assistants to retrieve and summarize incident protocols, aiding operators and reducing workload by 30%. Implemented predictive maintenance copilots using CNN crack detection on drone and LiDAR images integrated with generative retrieval. Ensured compliance with FedRAMP High and NIST 800-53 security standards. Established CI/CD pipelines for model drift detection and automated retraining.

Developed and deployed large-scale customer personalization and engagement ML models to segment users and recommend financial products. Implemented clustering models (K-Means, DBSCAN, Gaussian Mixtures) on multi-channel data increasing campaign response rates by 18%. Built hybrid recommendation engines using deep learning embeddings improving cross-sell conversions by 12%. Powered NLP-driven virtual assistants with BERT, spaCy, and custom intent classifiers, reducing live agent workloads by 22% and response times significantly. Automated deployment and retraining pipelines with AWS SageMaker, Docker, Jenkins, and MLflow ensuring scalable CI/CD. Worked closely with marketing and product teams to integrate ML insights into customer engagement strategies, boosting customer satisfaction by 15%.