Research Software Engineer
Rothamsted Research
Total years of experience :4 years, 10 Months
• Developed and optimised storage and accessibility for making agricultural research data publicly available including diferent modalities and sources for the First Agricultural Research Centre in the World.
• Researched state of the art practices in public data repositories for global research institutes and government organisations, and setup CKAN (Open Source Data Management) with its various functionalities such as automatic API serving, DOI minting, visualisations and preview using Python, Jinja templating engine and Bootstrap.
• Researched and implemented important meta-data handling practices for dynamic farm practices and applications, crop and livestock management in NodeJS based APIs to ship the North Wyke Farm Platform data.
• Studied and researched process based simulation models and utilised APIs to integrate them with Digital Twins for agriculture.
• Contributed to think tanks and workshops building the National Digital Twin Capability combining Process-based models, Artificial intelligence models and Emulators for crop yields, plant growth, sustainability, farming practices and net zero emissions including collaborations with the Alan Turing Institute, Imperial College London, and Exeter University.
• Studied and worked on the data recording and storage pipelines for experiments in Water, Soil, Plant, Livestock, Satellite and Weather in advanced agricultural research practice on the North Wyke Farm Platform at Okehampton.
• Regularly collaborated with Data scientists, Curators, Software developers, Managers and applied Agricultural and Computer scientists.
• Research and implementations in Multimodal Applied generative AI for Explainability in Medical Imaging Diagnostics, specifically for Chest X-rays and Brain images.
• Fine tuned in combination a stable diffusion model and a large language model (LLM) pretrained on radiology reports to generate corresponding normal counterparts of chest scans, affected with Viral Pneumonia, Lung Opacity and COVID, guided using natural language prompts.
• Examples of prompts include: Lung opacity on the left, Severe lung scarring on the top right, Healthy Lungs, Lung carcinoma etc.
• The discrepancy between the normal and abnormal gives rise to a diagnostically relevant map, visually explaining the models understanding of the condition similar to a radiologist.
• The model also exhibits emergent capabilities such as zero shot induction of unseen conditions e.g. Lung Carcinoma, Cardiomegaly, Tumour in healthy scans globally and with reasonable local geographical precision with controlled severity by hyperparameters.
• Publication under review in the Scientific Reports journal, Preprint available at Arxiv : VALD MD: Visual Attribution via Latent Difusion for Medical Diagnostics (arxiv.org).
• The work has also been validated as "State of the Art" in the field and has been selected for a chapter in the book called "Generative Machine Learning Models in Medical Image Computing" led by cutting edge research scientists from Oxford, Queen Mary University of London and others.
• Performed duties as a "Lecturer" in the Faculty of Science and Technology, in the Department of Computer Science for modules with content relevant to AI, Image processing, Computer Vision and Machine learning.
• Developed modules for Car detection, Number plate extraction and Super-resolution
• Built pipelines to access Google Street View data through APIs and analysed the data for information on local businesses.
• Developed modules for a website to display spatial data analytics for Hague.
• Conducted research on OCR, document modelling and digitising images of text documents to text.
• Developed business use cases for the use of AI in Law, Chat support systems, Financial markets and E-commerce
• Developed minimum viable products using AWS Kendra, Comprehend, Textract, Lambda and S3 APIs, combined with JavaScript.
• Made a proposal after extensive research which resulted in bringing NASA Scientist Dr. Radley Horton on board for collaboration on a tool to highlight climate change impact
• Led a team of interns to analyse the stock market and researched company valuation metrics to implement an automated web-based tool to help in investment analysis.
• Studied climate attribution, and spatial disaster data, visualised, and analysed the data and implemented dashboards to automate the analysis of such data.
• Studied and obtained satellite data using Google Earth Engine, specifically Landsat-8 and Sentinel-1 products for optical data and microwave backscatter, pre-processed the data using Band information
• Researched for factors which afect Crop yield, Live Fuel Moisture Content, Wildfire susceptibility and spread, extracted implementable use cases from research.
Developed a pipeline for displaying time-series analytics using TimescaleDB and Nodejs.
Developed modules on Angular for sharing information on different social media platforms and mail.
Developed modules of a mobile application using the Ionic framework.
Developed and integrated modules for large-scale management software using MVC ASP .NET, AngularJs, Bootstrap, HTML, CSS and Visual Studio team services.
Research and implementations in Multimodal Applied generative AI for Visual Attribution and Explainability in Medical Imaging Diagnostics