Aditya Bagwadkar

Leading R&D department with 7 team members to optimize and develop error-free scripts for products Novelty, Classifier and Patent QA by doing code reviews weekly and I conducted brain storming sessions for team members. Developed applications using Generative AI algorithm and used an LLM model fine tuned on clients custom documents to generate answers based on questions given by user from web page. Used AWS services to like AWS SQS, AWS S3 to handle users request properly when using Python APIs available through Django framework. Created an encryption based algorithm which discards the ability of LLM model to learn from confidential documents provided by clients. Used opensource models of Hugging Face to perform similarity search using FAISS indexing to get relevant documents based on user’s input text.

Developed applications using Stable Diffusion model and Controlnet to generate caricature faces of user. Fine tuned using Low-Rank Adaptation of Large Language Models to make Art. Designed a text to image editor tool using Fast API to generate images by providing prompts, negative prompts. Deployed AI system in GCP to generate Caricatures automatically using text prompt and Lang SAM model. Assisted members to design custom Q & A systems using Chat GPT , Claude API and Langchain.

Worked with the researchers of Northwestern Medicine to develop an AI product to help patients having cardiovascular disease to take necessary actions at the right time. Implemented algorithms in NLP and Computer vision to understand abnormalities present in blood pressure signals using time series data like ECG, PPG and spectrograms. Build a deep learning model architecture to clear the protocols provided by British Hypertension Society and Association for the Advancement of Medical Instrumentation. Under the guidance of data scientists, I learned to use time series data effectively and apply the knowledge in various sectors. Together with my team members and clients employees we deployed AI services using Microsoft Azure cloud services like AKS, Azure Web app service, Azure function app as well as used Dask , Flask frameworks. Used a reinforcement learning algorithm DQN agent, to identify the abnormalities present in blood pressure signals in hypertension and hypotension ranges.

Collected raw data and performed data engineering to build a good quality dataset. Under the guidance of a Senior ML engineer, I learned to use AWS cloud technologies for various production use-cases and practiced Github for version control, committing updates, making pull requests. Trained models in Amazon's EC2 instances and validate the model on a development set using metrics like precision, recall, iou coefficient, F1 score and accuracy. Learned to use FastAPI for deploying machine learning models as a service to the client. Experimented to train and deploy a model on Amazon EKS with Amazon Fargate and Amazon ECS with Amazon EC2 instance. Created Docker images and made available as Docker containers in Amazon ECR.

Hosted a website using WordPress, configured computation resources like VM, AKS in Azure. Prepared a project synopsis for disambiguating the problem statement, input/output format, defined test cases to understand at which scenario the solution works best and at worst, implementation method and feasibility. Applied transfer learning approach to apply classification objective in healthcare application. Improved the performance of model using ensemble learning of DenseNet-121, DenseNet-169 and DenseNet-203.

Collabaration with R&D team members to build a real-time media monitoring network that generates unique market data. Worked on spectographs and studied to identify and extract words from played radio songs containing noisy data like advertisements, radio commentary, etc. Applied Audio Content Recognition and Transfer learning technologies to recognize and extract the details like title of song, name of the artist. Extracted the total number of occupied seats and not occupied seats for data analysis which helped to understand which particular show is popular in the market using image processing algorithms. Used Natual Language processing algorithms like BERT, XLM-RoBERTa and other transformers to extract song titles , artist name based on song lyrics.

Developed value by creating AI system to localize metastasis regions of breast cancer from whole slide images. Worked on problems involving Whole Slide Imaging (WSIs) for tasks like segmentation and study of cancerous cells in Histopathology Images. Worked with different languages, platforms and frameworks such as Python, Numpy, Pandas, Matplotlib, Seaborn, Tensorflow, Keras, Pilow, OpenCV, Jupyter Notebook, Google Colab. Worked on processing DNA samples of viruses like MERS-COV, Rota virus, Zika virus, Ebola virus and used sequence models like Gated Recurrent networks, LSTM, RNN to classify genome sequences.