Akhil Kerhalkar

Currently working with the speech research team to build pipelines to automate machine learning model training and deployment on top of kube flow in AWS. Reduced the model training time devoted to keep the models updated with new data. Developed ingestion pipeline on top of Apache airflow with precursors such as format verification and data cleaning. Developed a library to serve audio and text data to AWS athena for fast querying and filtering.

Passed CCNA as a part of the internship and scored 920/1000. Developed Workforce management and Paid time off web tool to manage and schedule employees in shifts which decreased the time taken the managers to schedule and manage employee shifts. Built Linux from scratch (LFS) and learnt about various linux and network administration tools. Worked on internal Ansible playbooks to autodeploy test servers. Experimented on a POC on trying to use musl-libc instead of glibc on a read heavy in memory service and benchmarked IO performance.

Developed preprocessing pipeline capable of auto tagging data for performing NER using spacy which enhanced the preexisting NER module to work with minuscule data. Developed and fine tuned a model to extract checkbox data from unstructured data sources such as pdfs and images. This in turn helped enrich the OCR data being extracted.