Keshav Saraogi

Results-driven IT professional with hands-on experience in building enterprise AI solutions and scalable backend architectures. Proficient in developing RAG systems, fine-tuning LLMs, and designing data pipelines using tools like LangChain, Airflow, and FastAPI. Demonstrated success in microservices deployment, CI/CD optimization, and API engineering. Adept at solving complex data and system integration challenges in fast-paced environments.

Artist classification and recognition using big data and deep learning

• Designed a scalable audio ingestion and preprocessing pipeline to handle 10GB+ of raw audio data using the Hadoop Distributed File System (HDFS) and Apache Spark on AWS EMR, optimizing for the efficient preparation of high-volume datasets for ML tasks
• Developed modular ETL workflows with PySpark and Spark MLlib to extract audio features (MFCCs, Chroma, Spectral Contrast) from WAV/MP3 files, transforming them into time-frequency spectrogram matrices, resulting in a 30% reduction in data preparation latency
• Automated the end-to-end data processing lifecycle using AWS Step Functions, integrating raw data retrieval, feature extraction, and S3-based storage with prefix-based partitioning and versioning to enhance auditability and scalability

Object detection pipeline on AWS

• Engineered a complete image data pipeline for object detection using AWS S3, SageMaker Processing Jobs, and OpenCV, supporting real-time ingestion and preprocessing of large-scale annotated datasets
• Implemented robust data augmentation and annotation standardization workflows using Pandas and Boto3, reducing preprocessing time by 40 percent and ensuring consistency across diverse metadata formats
• Built ETL components for bounding box extraction, format conversion (Pascal VOC ↔ YOLO ↔ COCO), and dataset partitioning (train/validation/test) using Python and Dockerized SageMaker scripts for deployment at scale