* Leading Product Development for our product "DocAI", achieving sub-second document processing latency.* Explored applications of LLMs for guided information extraction and conversational interfaces, including chatbot development. Conducted comparative evaluations of LLMs against existing deep learning solutions, analyzing accuracy, latency, and cost-effectiveness, leading to identification of high-impact use cases for integration into internal products.* Led project to replace commercial form recognizer with in-house ML solution, improving F1 score by 8% and cutting inference costs by 85%.* Designed tools for standardizing data extraction model training, reducing cycle time by ~40%.

- Built performance tracking solutions for end-to-end assessment of ML Pipelines, aiding in early detection of model quality degradation .- Built Kubernetes cluster for our SaaS solution, handling 20x previous throughput with similar per page average processing cost.- Mentored ML interns, guiding critical POCs into Production.- Revamped model inference architecture for efficient on-demand serving of multiple models from a single container, reducing memory footprint by 30-40%.

- Automated business processes in production setting using Machine Learning/Deep Learning, manage Machine Learning model lifecycle, build custom ETL pipelines according to business needs.- Worked on wide range of NLP tasks including Classification, Named Entity Recognition, Extractive Question & Answering on domain-specific Visually Rich Documents. Worked on adapting state-of-the-art research methodologies, dealing with Visually Rich Documents for our use-case with great results.- Designed internal model performance tracking solution custom to the company's end-to-end workflows for better model performance and lifecycle management.- Designed feedback loop systems ( integrated into our SaaS product ) for continuous feedback and improvement of our models.

• Using Deep Learning to perform edge detection (Holistically Nested Edge Detection) in order to detect boundaries of agricultural plots from satellite imagery• Automate the process of downloading satellite imagery from Landsat 8 and analysis ofgiven agricultural plots • Digitization of Identity Documents using OCR, image pre-processing and then using both logic and automation to extract the information from those cards in a structured format