Siva Sai Srinivas Gurrala

Automated Alt-text generation using BLIP(Support LLMs). Fine-tuned BLIP on a custom image dataset and established a robust pipeline for generating alt-text, short, and long descriptions for PDFs and e-books, utilizing LLAMA 2 70B, GPT vision 4, and GPT 3.5 turbo for description generation. Developed and implemented an efficient client support chatbot utilizing the RASA framework. Specialized in technical support utilizing an LLM (LLAMA 2 7B) fine-tuned on the internal platform (DigiEdit). Finetuned T5 encoder-decoder model from HuggingFace on a large set of in-house GEC sentence pairs.

Collaborated on the construction of high-throughput Deep Learning servers, leveraging Nvidia RTX 3rd Gen GPUs to achieve the capability of processing 10,000 images per second. Designed a pipeline that crops, rotates the improper and improves the DPI of the documents. Pioneered and executed an optimized Machine Learning pipeline, achieving a remarkable 95% accuracy in printed document data extraction and 72% accuracy in handwritten document data extraction using DOC AI and Google Vision.

Spearheaded the creation of a video compression and processing pipeline in AWS through the development of a custom Python script utilizing the Boto3 library. Played a pivotal role in Fintech initiatives, crafting solutions using the PDFtk library to showcase data sourced from AWS S3 buckets. Architected APIs using the PHP Slim framework to dynamically address runtime queries and data retrieval. Led the development of a mobile application for the Andhra Pradesh Police Department using the Ionic Framework.