Sandeep Polavarapu Email & Phone Number

New York, Us

Chicago, Il, Us

• Designed Generative AI applications using HuggingFace open-source LLMs, GPT-4 Vision models, and LangChain to build a multi-modal question-answering application.

Leverkusen, North Rhine-Westphalia, De

- Developed Generative AI applications using OpenAI GPT foundation models and advanced RAG models using Langchain and Pinecone Vector databases and streaming outputs to reduce latency.- Built customized AI Agents using function calling and integrated various services and APIs leading to an overall 15% reduction in operational man-hours.- Collaborated globally across NA, EMEA, LATAM, and APJ under tight timelines, working alongside diverse domain experts such as breeders, lab scientists, and data engineers, adeptly handling the complexities of catering to each stakeholder's unique requirements.- As a Data Scientist within the Analytics and Product Design (APD) team, positioned under the Crop Breeding Department, I leverage my expertise in data and business-driven methodologies during the pre-commercial stages of our initiatives.- Upgrading, supporting, and maintaining the Inbred EVA project. This advanced crop model, built using data-driven and business-aligned strategies, is designed to provide optimized recommendations to farmers during the crucial pre-commercial phase. This ensures more informed seeding and planting decisions, fostering enhanced productivity in the future stages.

Detroit, Michigan , Us

- Led the MLOps initiative by implementing ready-made Sagemaker pipelines for some teams, and customized solutions using tools like Docker, Kubeflow, GitHub actions, fast API, and airflow ensuring end-to-end integration including tests, scalable training and deployments, monitoring, and continuous learning reducing in Time to Market (TTM) of new models by 25%.- Developed a modularized computer vision Generative AI pipeline using pre-trained ControlNet (Stable Diffusion) that enables users to generate a redesigned room from an image and text prompt using Python and HuggingFace achieving an FID score of 15.5 and a 20% increase in user engagement.- Built a Lookalike finder using computer vision techniques to assist users in finding similar-looking furniture products at a lower price scrapping various websites and APIs, SQL, and NoSQL databases deploying on AWS Sagemaker resulting in a user satisfaction rate of 86% and a 35% reduction in average customer spending

• Developed Time series crop classification on multispectral satellite images from large data sources (>10TB) using stacked Bi-directional LSTM to study CO2 emissions of sugarcane burning in Thailand, achieving a producer accuracy of 96% and user accuracy of 92% which was 8% above the baseline.• Implemented a Computer Vision Semantic Segmentation model using DeepLabV3+ to map burnt areas in sugarcane residue images, resulting in an IOU (Intersection over Union) score of 77.7% and presented the findings of the above study as a report at the NASA LCLUC conference. • Devised a statistical model using SciPy numerical optimization on soil and weather data for winter wheat Phenology dates prediction. This was used to make logistic and farming decisions.

Cambridge, Ma, Us

• Developed and evaluated Anomaly Detection ML pipelines using A/B testing on Decision Trees, SVM, XGBoost, CNN-LSTM, and Autoencoders to detect and mitigate DDOS (Distributed Denial of Service) attacks. • Deployed models with Docker, AWS ECR, SageMaker, Apache Spark (PySpark) and Kinesis streaming. Monitored data drift and performed continuous training, reducing critical event reaction time by 30% and achieving an 82% recall score.

Cambridge, Ma, Us

• Engineered ETL pipelines, developed and deployed a secure NLP-based Topic modeling Document Information Retrieval system using LDA (dimensionality reduction), K-Means clustering, and knowledge graphs on hybrid cloud (AWS S3). • This resulted in (a) a Semantic Similarity score of 0.71, (b) a MAP score of 0.5, and (c) an Increase Click Through Rate of 0.1.Tools: TensorFlow, AWS SageMaker, PySpark, GlueETL, Confluence, Jira, MongoDB, MuleSoft, API, CI/CD Jenkins, Tableau, GCP

Cambridge, Ma, Us

Developed a PoC (Proof of Concept) and ETL pipelines for JIRA to Salesforce system integration using Mulesoft (integration bus) and Oracle Database.Tools used: Salesforce Apex, Mulesoft, SOQL, ODBC, JIRA.

I was a part of the image analytical team for PISAT (PESIT Imaging Satellite), an indigenous satellite developed by PES University and launched by ISRO (Indian Space Research Organization). Utilized super-resolution and standard image processing on RGB and near-infrared image data to obtain the analysis of remote sensing, soil health detection (vegetative use and cover), and weather analysis using R.Tools Used: Python, R, OpenCV

Master'S In Professional Studies, Machine Learning

Bachelor Of Technology - Btech, Computer Science And Engineering

Major: Bachelor Of Technology (B.Tech.), Computer Science & Engineering