Athira B

Experience

• 

  Tata Consultancy Services

  Nov 2015 - Jun 2021

  DataStage Developer | ETL DeveloperRaw data from different sources are collected and transformed based on the business logic to provide insights to stake holders.Demonstrated team leadership - coordinated Dev and Production team for an ahead-of-schedule DataStage version upgradeBuilt ETL jobs to create pipelines for data extraction to processing to transformation and loading to the Databases. DataStage Developer | ETL Developer Talend Developer | ETL Developer Oracle Data Integrator Developer | ETL Developer

  • Information Technology Analyst

    Nov 2020 - Jun 2021

  • System Engineer

    Nov 2017 - Nov 2020

  • Assistant System Engineer

    Nov 2016 - Nov 2017

  • Assistant System Engineer - Trainee

    Nov 2015 - Nov 2016
• 

  Tiger Studios

  Nov 2021 - Nov 2022

  Co-Founder: Product and Marketing

  Game Ideation and Design:I actively participated in the conceptualization and design phase of our gaming projects. This involved brainstorming and formulating innovative ideas, defining game mechanics, and overseeing the overall design process to ensure a compelling and engaging user experience.User Sentiment Analyzer using NLP:Recognizing the importance of understanding user feedback, I implemented a User Sentiment Analyzer using Natural Language Processing (NLP). This technology enabled us to gauge user sentiments expressed in reviews and comments, providing valuable insights into user satisfaction and areas for improvement.Market Research, Survey, and Focus Groups:To inform our decision-making processes, I conducted comprehensive market research initiatives. This involved deploying surveys and organizing focus groups to gather qualitative and quantitative data. These efforts were instrumental in understanding the preferences and expectations of our target audience, shaping our game development and marketing strategies accordingly. Show less

• 

  SupportVectors

  May 2023 - Dec 2023

  Data Scientist

  Medical Assist AIDeveloping AI components for a prototype to make doctor-patient interactions more seamless. Now, doctors can focus on patients rather than worrying about the extensive documentations and procedures that they have to follow post consultation. The doctors can record conversations with the patients and generate transcript with speaker diarisation, SOAP format medical summary notes, follow-up questions and ICD (International Classification of Diseases) codes - All in a matter of few seconds and click of a button!Key components: Automatic Speech Recognition + Speaker Diarisation, LLM, Few-Shot In-Context Learning, RAG, Vector Embeddings, Google Healthcare APIMulti-Modal Search EngineDeveloping the AI components for an enterprise grade vector database and search application. This vector database has the hybridized power of semantic and lexical search along with the magic of LLMs. This multimodal product can handle text, images, audio and video files. A relational database is used to store the relevant information needed for retreival.Key components: Parsing, Semantic Chunking, Text Embeddings, BLIP, ASR, Multimodal Embedding - ImageBind, FAISS + HNSW Indexing and Similarity Search, Elastic Search, Cross Encoder - Reranking, RAG, LLM, Postgress DBAI AuditCreating SOPs to audit clients’ AI infrastructure.Developing a knowledge portal for educating the clients regarding thesame. Show less

• 

  Stealth Startup

  Jan 2024 - now

  Data Scientist

  Text Classification - Careful curation of custom datasets for training and testing to get the best possible model - Fine tuning BERT based models (Distilbert, Deberta) to increase the accuracy of a pretrained model (24%) to 85%+ accuracy in real world applications - KNN based classifier (K nearest neighbors) Image Classification - Careful curation of custom datasets for training and testing to get the best possible model - Trained YOLO on custom data with 92%+ accuracy in real world applications - Fine tuned ViT to increase the accuracy of a pretrained model (18%) to 80%+ accuracy in real world applicationsLLMs (Large Language Models) - Data labeling - Data augmentation - Synthetic data generationText Extraction from multi modal data- Document AI pipeline for document parsing and loading to BigQuery- Custom text extraction pipeline from documents, audio, video.- Text extraction from images using OCR- Employing different chunking strategies- NER models for PII, PHI. Implementing custom detection patterns.Approach- Each task is performed after careful research- EDA, comparative studies, detailed data visualizations Show less