• Used Kubeflow to build pipelines for speech recognition model benchmarking, enabling data scientists to test their models in a matter of hours instead of days.• Developed a Cobra CLI tool using Go Routines to build parallel audio preprocessing pipelines for large datasets (300 GB) on AWS S3, reducing manual efforts in model training setup by over 30%.• Wrote an A/B testing library using Python, SQL, Metabase, and SQLAlchemy, increasing resolution rates by 20% across 3 clients while… Show more • Used Kubeflow to build pipelines for speech recognition model benchmarking, enabling data scientists to test their models in a matter of hours instead of days.• Developed a Cobra CLI tool using Go Routines to build parallel audio preprocessing pipelines for large datasets (300 GB) on AWS S3, reducing manual efforts in model training setup by over 30%.• Wrote an A/B testing library using Python, SQL, Metabase, and SQLAlchemy, increasing resolution rates by 20% across 3 clients while abstracting complex join and window queries. Show less

Refactored Stride's Multilingual Keyphrase recognition API to work with Spanish. Performed Load Testing and Functional Testing to benchmark the API. Diminished API error rate by 30 percent and overhauled API RPM by 50%.

Implemented SQL and Object Oriented ABAP to devise bank specific regulatory services for international payments.Wrote a white paper for Knowledge documentation and dealing with similar use cases.Built a MVC Web App for monitoring retail data using SAP UI5 and Javascript.Rewrote reporting interfaces and queries to perform OLAP using OpenSQL and S4/HANA.

Designed a general purpose framework to train, retrain and deploy voice assistants. Built an API testingdashboard to record response times and identify bottlenecks in several text and speech processing workflows. Deployed this framework as an API on AWS EC2. Reused intent identification and entity extraction models using NLTK and sklearn, and deployed these as lambda functions on AWS.

Used Selenium, multiprocessing and Tesseract to build one of the largest repositories ofIndian names on the web, that would then serve as a lookup table for identifying Name specificinformation in end-user chat messages. This helped the Operations team easily identify verified,real-world end users, drastically reduce noise in internal user databases caused by spam andnon-contributive end users/end user messages, and reduce investment in external 3rd partyverification… Show more Used Selenium, multiprocessing and Tesseract to build one of the largest repositories ofIndian names on the web, that would then serve as a lookup table for identifying Name specificinformation in end-user chat messages. This helped the Operations team easily identify verified,real-world end users, drastically reduce noise in internal user databases caused by spam andnon-contributive end users/end user messages, and reduce investment in external 3rd partyverification services.Refactored the chat validation module that is used to categorize and validate incoming end userchat messages into one of many different categories, making use of Syntactic and Morphologicalfeatures. Increased recall of the same from 85% to 95%.Built an Aadhar verification module to exploit QR code encoded end-user info rather than VanillaOCR, decreasing reliability on 3rd party OCR engine by 33%. Show less

*Built a RF model to classify passages of text from low resource datasets into specific topics using scikit-learn, NLTK and genism. Heavily made use of Feature Engineering and Feature Ranking techniques like LDA and PCA to find best-fit features for the same dataset.*Contributed to ongoing NLP research, specifically in the domains of Information Extraction, Information Localization, and Natural Language Understanding.*Built Object Detection/Classification models for intelligent… Show more *Built a RF model to classify passages of text from low resource datasets into specific topics using scikit-learn, NLTK and genism. Heavily made use of Feature Engineering and Feature Ranking techniques like LDA and PCA to find best-fit features for the same dataset.*Contributed to ongoing NLP research, specifically in the domains of Information Extraction, Information Localization, and Natural Language Understanding.*Built Object Detection/Classification models for intelligent extraction and localisation of PDF entities. Show less

Constructed an UAV capable of navigation and data capture/processing. Designed Image Processing pipelines and networking infrastructures for data capture, transmission and preprocessing using OpenCV and Bash.