Projects

Intelligent Exploration of 3D Printing of Pharmaceuticals using Artificial Intelligence

Partnered with Dr. Rajkumar Velu from IIT Jammu to scale up manufacturing resources and implement an end-to-end automated process for solid pharmaceutical production, leading to a 50% reduction in production time. Developed a cutting-edge technology for solid pharmaceutical production covering all the aspects namely formulation design, drug prediction, drug administration, and Quality Control check using the power of Artificial Intelligence Algorithms....

Tech Stacks: Tensorflow, Pytorch, Keras, Scikit-learn, Python

Satellite Imagery Object Detection using Dual Backbone Architecture

The model incorporates Convolutional Neural Networks and Regional Proposal Networks to achieve high accuracy in object detection. Developed a prototype to detect objects such as Buildings, water bodies, trees in Multi-scale Satellite Images, turing to be an advanced version in conjugated neural networks.

Tech Stacks: Tensorflow, Pytorch, Scikit-learn, Roboflow Annotate, Python

AdCNNQTEnsemble - Medical Image Diagnosis using Advanced CNN feature and DAP Analysis

Developed and implemented AdCNN, a cutting-edge model for Medical Image Diagnosis that achieved unprecedented accuracy rates of up to 93% in lab experiments. Achieved accuracy improvements ranging from 86% to 93% across four different diseases, by collaborating with Andijan State Medical Institute to evaluated AdCNN’s performance on real-world datasets.

Tech Stacks: Tensorflow, Pytorch, Keras, Scikit-learn, Python

Handwritten Text Recognition using Robust Sequential Learning Approach

Developed and implemented a state-of-the-art handwritten text recognition model using modified EfficientNetV2, Stacked RNN, and BiLSTM architectures, resulting in a 15% improvement in accuracy compared to previous models. Conducted extensive analysis of the developed model by evaluating its performance on various benchmark datasets, achieving an average precision of 95% and a recall of 93%, outperforming existing state-of-the-art models.

Tech Stacks: Tensorflow, Pytorch, Keras, Python

False Virtual Wallet Detection and Analysis using LSTM and CLSTM

This project aims to detect 11 types of false Crypto keys using LSTM and Conjugated LSTM models. The models were trained on a large dataset of Crypto keys, and the results showed that the LSTM model achieved a final accuracy of 90%, while the Conjugated LSTM model had an accuracy of 86%, which is better than the other models suggested. The models were also used to forecast the future volume of Crypto transactions. The results showed that both LSTM and Conjugated LSTM models were able to accurately predict the future volume of Crypto transactions, based on historical data.

Tech Stacks: Tensorflow, Scikit-learn, Python

Recommendation Systems using AWS Factorization Machine and other MXNet Models

Engineered a cutting-edge recommendation system using MXNet and AWS SageMaker Factorization Machines Algorithm to analyze user preferences and provide personalized movie recommendations, achieving an accuracy rate of 87%, and compared the accuracy rate with other custom Machine learning models.

Tech Stacks: Tensorflow, AWS SageMaker, MXNet, Python

Object Detection in AWS SageMaker - A Quick Overview

Developed and optimized an Object Detection model using AWS SageMaker built-in algorithm, resulting in a 20% improvement in accuracy compared to previous models. Implemented an automated workflow using AWS Ground Truth for labelling objects, reducing the time required for data annotation by 50%.

Tech Stacks: AWS SageMaker, AWS GroundTruth, Python

ChatBot using LLM - Powered By LangChain - Document ChatBot

Developed and optimized an Object Detection model using AWS SageMaker built-in algorithm, resulting in a 20% improvement in accuracy compared to previous models. Implemented an automated workflow using AWS Ground Truth for labelling objects, reducing the time required for data annotation by 50%.

Tech Stacks: AWS SageMaker, AWS GroundTruth, Python

Event Driven Pipeline to Summarize the Meeting Transcripts using Llama3

Developed an automated system for transcribing audio files and generating summaries using AWS services. Implemented functionalities to handle audio file upload, transcription job creation, content splitting based on speaker labels, and summary generation. Managed the setup of an S3 Bucket and integrated it with the AWS Transcribe service for audio file storage and transcription. Utilized Boto3 client to automate the creation of transcription jobs, reducing manual intervention. Successfully reduced the time required for transcription and summary generation processes, enhancing productivity. Improved workflow efficiency by automating manual tasks, resulting in time savings and reduced operational overhead. Enhanced accuracy in content splitting and summary generation, providing high-quality output for stakeholders.

Tech Stacks: Python, S3, Transcribe, Bedrock, LLM, GenAI, Lambda