Janvi Pandya

• Created a product recommendation system using Azure Synapses FABRIC and a collaborative filtering model. • To visualize the predictions used two methods, the first being PowerBI and the second being .NET’ Blazor component • Created a Product Recommendation System using Azure FABRIC Synapse. The model used for the same was a collaborative filtering model. The recommendation system was created with the help of Pyspark, MLflow and the predict() function of MLflow to write the recommendations back to the Lakehouse. • To visualize the predictions two methods were used, the first being PowerBI and second being .NET blazor component. • Worked on a computer vision project of Species Detection in real time using YOLO. • Created a warranty check chat bot using RAG and OpenAI on a vectorDB • Demonstrated dedication to continuous learning and improvement by attending Microsoft Partner events, workshops and getting certifications. • Strengthened team cohesion through regular communication updates, progress reports and constructive feedback sessions.

Project 1: Healthscore • Leveraged Azure SQL and Cosmos databases by utilizing SQLAlchemy and PyODBC to extract data, apply logic, and update the databases effectively. • Conducted exploratory data analysis to gain insights and understand underlying patterns. • Performed pre-processing tasks such as handling missing values, duplicates, and balancing the data. • Developed a robust Dutch language model using BERT for grammatical error correction, achieving an accuracy rate of 84%. • Designed a hybrid tokenizer capable of tokenizing paragraphs at the sentence, word, and character levels, depending on the specific requirements. • Utilized T2T and LSTM models for prompt engineering to create an AI solution. • Orchestrated the deployment of the project on Azure Kubernetes Service through Docker, streamlining automation processes. • Constructed a RESTful API using FastAPI, enabling seamless data extraction, transformation, and loading into the database. • Integrated ChatGPT into the project to provide clients with valuable insights and recommendations.

Project 1: Prestigious Client Projects (Rolls Royce, Airbus, Ford) • Successfully executed projects for esteemed clients suchas Rolls Royce, Airbus, and Ford. • Developed a green project for a client involving the detection of palm oil plantations in Nigeria using satellite imagery provided by the client. • Utilized the Selenium web-driver tool to automate the process of capturing screenshots from satellite map images based on latitude and longitude inputs saved in an Excel file. • Implemented the VGG16 model of Convolutional Neural Network (obtained using Keras) and OpenCV to detect palm oil trees. • Employed a test:train split of the dataset [20:80] and created an interactive UX|UI interface that allowed users to search and detect palm trees within a specified area on the map of Nigeria using a bounding box obtained via the Google Maps API. Utilized Pix-to-pix GAN to create a mask over the detected palm trees, calculating the plantation area in the selected region. • Utilized the SNAP (ESA) tool and a QGIS plugin for pre-processing multispectral images, incorporating k-means classification within the QGIS plugin. • Integrated JavaScript UI code with the backend Python code using the Flask framework. Project 2: Augmented Reality Model of Jet Engine Cross Section • Deployed a three.js-based 3D model showcasing the cross-section of a jet engine, viewable in augmented reality on both iPhone and Android systems. Also, Enhanced the model by incorporating dynamic functions simulating engine operation and adding interactive hotspots Title: Use of Augmented Reality and Virtual Reality in the Healthcare Industry [Royal Holloway University, 2022] • Along with an in-depth literature review a quantitative data analysis was made with the help of SPSS on a dataset that was obtained with the help of a survey • The survey was designed and circulated with the help of Qualtrics and a total of 411 responses were obtained • The data was then analyzed with the help of statistical tests