Sidharth Pal

- Automated Product-Cataloguing with Efficient-Neural architectural search and Multi-Task Representation learning on multiple product label granularities. - Achieved 8% better results than existing solution with less than half development time. - Speaker at the SAP International Conference on Machine Learning CML 2021. - Contributed to a pinned feature-request on Auto-Keras GitHub project for image feature extraction.

- Implemented Catalog-Search – a Content-based Image retrieval engine meant for industrial product catalog with. - Overcame the common feature collapse issue with proposed sigmoid triplet sequencing approach for Deep metric learning using Keras/TensorFlow with Data Augmentation. - Containerized deep learning workflows on Google Cloud Platform (GCP). - Achieved 92% top10 accuracy for datasets with less than 1000 categories.

- Automated resource allocation and deployment (job) scheduling workflows by curating Database response time outlier detection and Server metric load forecasting (variance estimation) pipelines. - Saved 10+ workhours per week by orchestrating various CI/CD and Data Management pipelines.

- Spear-headed a team of 5 members for project-wide implementation of automated configuration push and vertical scaling workflows, driven by Kibana Machine learning bucket analysis on multiple data sources. - Slashed server failures by 85% with application downtime prediction and automated recovery pipeline implemented using ITSM and event management data.   - Automated troubleshooting and remediation of client-side issues with a Natural Language Processing client.

- Managed multiple Big Data (Hadoop) clusters, with capacity planning and HDFS setup. - Saved 20+ work hours per week, by enabling automated central server administration and assisted root cause estimation with a data driven web portal (Java/Ansible).

Focus: Deep Learning and Computer Vision, Non-Relational Databases, Advanced programming principles (Java & Python) Summer Project: - Visual-to-Audio Aid for visually impaired – A system to process visual input on timed frames to produce sensible audio aid in accordance with human information processing limits, using image captioning, semantic text comparison and text-to-speech modules. GitHub Link: https://github.com/sidphbot/visual-to-audio-aid-for-visually-impaired

Focus: - Microprocessor and Microcontroller - Multivariable Calculus & Linear Algebra - VLSI - Data Structures