I'm Rishabh Singh, a passionate and skilled AI/ Robotics Engineer with a strong foundation in C++/Python, Machine Learning, Deep Learning, Computer Vision, LLM, Robotics and MLOps. Currently pursuing Master of Science in AI/Robotics at Northeastern University gaining hands-on experience in cutting-edge technologies and algorithms. I'm eager to leverage my expertise to solve real-world challenges. Let's connect and explore how we can collaborate to push the boundaries of technology.
Introducing myself in a single sentence : "I am an Ordinary man with Extraordinary dreams".
Sep 2022 - Dec 2024
Robotics & AI
GPA: 3.91/4.00
Jun 2016 - Jul 2020
Electronics and Telecommunication Engineering
GPA: 8.23/10.00 (Graduated with first class with distinction)
Jul 2023 - Dec 2024
Developed a cutting-edge object detection model using ResNet50 with Faster R-CNN and MobileNetV3 on HPC, optimizing for real-time performance with PyTorch and CUDA. Managed a 256 GB dataset of six classes and led data preprocessing, converting raw images to 16-bit TIFF format through debayering and transforming to linear and log spaces for accurate model evaluation.
Aug 2023 - Dec 2024
Supported new international student's cultural transition and academic success through one-on-one and group meetings, fostering an inclusive community and promoting intercultural competence.
Jan 2024 - Aug 2024
Implemented advanced visualization and contour-based segmentation for shape transformation assessment, automating the annotation of biomedical shapes into four categories. Utilized Hu Moments and XGBoost for classification, achieving high accuracy in correlating shape changes with blood pressure.
Employed Vision Transformer and LSTM models to capture spatial and temporal aspects, while developing algorithms for dual IMU sensor data and creating a dashboard for visualizing analysis. Enhanced model interpretability using SHAP and refined protocols for improved detection of cardiovascular events.
Jan 2022 - Mar 2022
Created a distracted driving detection model using the State Farm Dataset, achieving 98% accuracy with CNN and 99% with ResNet-101 through transfer learning and hyperparameter tuning. Utilized feature extraction and visualized results with Class Activation Mapping (CAM).
Utilized 5 feature extraction techniques: HOG, SURF, Color Histograms, and applied PCA for dimensionality reduction
Sept 2021 - Dec 2021
Engineered a wireless surveillance bot utilizing YOLOv4 for object detection, achieving an accuracy of 92% across different areas and 92.17% at different times of the day.
Jun 2019 - Aug 2021
Developed an end-to-end ML pipeline for sentiment analysis using LSTM networks, achieving 89% accuracy, and a predicting API for real-time sentiment classification. Implemented CI/CD using CircleCI and AWS for scalable deployment and automated updates of the best-performing model.
Parsed over 100 resumes using NLTK for Named Entity Recognition, achieving an F1-score of 0.89 in predicting job roles with Random Forest and Gradient Boosting models through skill extraction and feature transformation.
Extracted resume skills and applied supervised learning random forest, gradient boosting with 0.89 F1-score to predict job.
The RAG-based Flight Query Bot is a Streamlit-powered web application that allows users to query flight information using natural language (e.g., "Show me flights from New York to London"). It uses Retrieval-Augmented Generation (RAG) with an Ollama language model to extract entities from queries and retrieve flight data from a mock database.
The AI Agent is a Streamlit-based web application that automates machine learning workflows by allowing users to upload datasets, select targets and features, and train models using an initial scikit-learn pipeline. It leverages an Ollama-hosted LLM to proactively generate Python scripts, displays performance metrics and plots aiming to simplify ML tasks for users while providing reusable scripts.
Pioneered Generative AI Chatbot leveraging RAG with GPT, LLama 2, and Flan-T5 LangChain, HuggingFace, and Chroma to extract insights from 10-Q reports. Built and deployed the chatbot on Docker using Dash, enabling semantic search, summarization, and key performance indicators (KPIs) extraction from financial documents.
Fused Hugging Face pre-trained tokenizers, Vision Transformer (ViT), and LLMs, achieving 0.29 WUPS by experimenting with (ViT + BERT) encoder for classification and (ViT + BERT) encoder + (GPT-2) decoder for answer generation.
This project focuses on developing an affordable and user-friendly eye-gaze controlled wheelchair to restore mobility and independence for patients suffering from ALS. The system utilizes gaze tracking technology to interpret eye movements and translate them into wheelchair navigation commands.
Built a navigation stack using two different sensors - GPS & IMU, understand their relative strengths + drawbacks, and get an introduction to sensor fusion.
The project's main goal is to investigate real-time object detection and tracking of pedestrians or bicyclists using a Velodyne LiDAR Sensor. Various point-cloud-based algorithms are implemented using the Open3d python package. The resulting 3D point cloud can then be processed to detect objects in the surrounding environment.
This project involved developing an autonomous system using mobile robots for disaster response. The system generated a complete map of an initially unknown environment and located any victims using AprilTags. Off-the-shelf components were used, and explore lite was modified to improve performance.
The main aim of the project is to implement Content Based Image Retrieval which is one of the most important concepts of computer vision. The bog companies are using this to compare the similarity between the images based on the feature vector of the Targeted image.
This project is about real-time 2D object recognition. The goal is to have the computer identify a specified set of objects placed on a white surface in a translation, scale, and rotation invariant manner from a camera looking straight down. The computer should be able to recognize single objects placed in the image and identify the objects.
This project is about learning how to calibrate a camera , then use the calibration to generate virtual objects in a scene. After getting calibration parameters System will be able to identify a target and then position a virtual item in the scene next to the target so that it moves and orients itself appropriately in response to camera or targets.
Lane detection is a crucial step in training autonomous driving cars. It helps identify and avoid entering other lanes by analyzing visual input. Lane recognition algorithms play a vital role in ADAS and autonomous vehicle systems. They accurately detect lane locations and borders, contributing to safe and reliable navigation.
Implemented a 5x5 Gaussian filter, 3x3 Sobel X and 3x3 Sobel Y, generated a gradient magnitude image from the X and Y Sobel images, blurred and quantized a color image, did live video cartoonization, put sparkles into the image where there are strong edges.
Conducted an in-depth data analysis for a hotel chain. I began by thoroughly understanding the core business challenge and then leveraged a Kaggle dataset. My work encompassed data cleaning, transforming the dataset, and extracting valuable insights to provide a comprehensive understanding of the hotel chain's data.
Assembled a Tortoise bot from scratch and am currently programming it to generate a complete map of an unknown environment. The project involves implementing SLAM (Simultaneous Localization and Mapping), Explore Lite, and occupancy grid mapping for ROS2-based applications.
Visual Odometry and Structure from motion (SfM) Pipeline for 3D Reconstruction and Pose Estimation. Working on integrating KITTI dataset, OpenCV, ORB/SIFT features, FLANN/BFMatcher, RANSAC, and essential matrix for pose estimation to achieve 3D reconstruction, trajectory construction, triangulation, bundle adjustment, scene visualization using GTSAM.
Swipe tabs to explore more categories.
Below are the details to reach out to me!