About

Hands-on Project:

SELLER-CENTRAL AI-CHATBOT SYSTEM

• Developed an AI-powered chatbot system that delivers real-time, actionable insights to Amazon sellers by analyzing sales data, customer reviews, and product metadata.

• Designed and implemented a scalable data pipeline on GCP for ingesting and preprocessing Amazon seller data—leveraging BigQuery, cloud storage, and advanced techniques (Handling missing values, invalid values, duplicates, feature extraction, normalization, recursive chunking) to transform raw user reviews and product metadata into structured data; engineered vector embeddings using Hugging Face transformers and indexed them in ChromaDB/AstraDB for efficient similarity-based retrieval.

• Integrated ML models like Sentiment Analysis using NLTK libraries, SVM, Random Forest into an end-to-end RAG pipeline for feature extraction and real-time insights—built sentiment analysis models with GPU acceleration and batch processing, orchestrated data retrieval and prompt construction using LangChain, and deployed the solution with FastAPI, Docker, and LangSmith monitoring to ensure reliability and scalability in production. 

• Employed significance tests, such as the t-test, chi-square test, and ANOVA, to analyze features, identify bias, and assess model performance through comparisons of metrics like accuracy and response time.

• Evaluated ML models using multiple evaluation metrics including accuracy score, r2 score, MSE, MAE, classification reports; Collaborated with a cross-functional team to establish CI/CD pipelines and implement multi-agent workflows, resulting in a reliable, end-to-end MLOps chatbot system for e-commerce analytics.

SEMANTICALLY GUIDED FEATURE MATCHING FOR VISUAL SLAM

• Developed and implemented localization accuracy in SLAM pipelines by integrating semantic descriptors with ORB features to enhance feature matching using Python and C++ and reduce drift, and better shape estimation, achieving a 20% reduction in translation errors on KITTI datasets.

• Evaluated the system on large-scale datasets (KITTI, P4B) using ORB-SLAM2, PyTorch, EVO Python Package, achieving significant improvements in translation accuracy (reduction by 2-3 meters), 3D reconstruction using map points and loop closure robustness by leveraging pre-trained segmentation models (PIDNet).

• Examined performance on benchmark datasets using EVO metrics, and demonstrated proficiency in SLAM concepts, feature-based visual odometry, and classical geometric computer vision while contributing to a modular codebase. 

OBJECT DETECTION USING YOLO V8 AND COMPUTER VISION TOOLS 

• Developing a model for object detection with higher precision and accuracy using YOLO V8 

• Collected the data, annotated the collected data, and formatted the data structure to input into YOLOv8 using the CVAT tool. Trained a machine learning model (object detector) using YOLO v8 nano model for 100 Epochs 

• Currently testing the trained model using various unseen datasets including video processing to analyze its performance and accuracy in predicting the model using Computer Vision tools

UNMANNED AERIAL VEHICLE NAVIGATION SYSTEM

• Developing a hybrid navigation system combining Q-Learning and computer vision for autonomous drone control

• Achieving an 80% collision avoidance rate using real-time obstacle detection while considering its dynamics and control systems and reduced navigation steps by 55% through reinforcement learning as of now

• Integrated multiple camera systems for comprehensive environmental awareness and path optimization

PARKING SPOT DETECTION AND COUNTER | COMPUTER VISION

• Developed a real-time system using Python and OpenCV to detect parking spots and determine occupancy status based on video feed analysis, leveraging connected components analysis and pre-trained ML models - GitHub.

• Designed visualization outputs with bounding boxes (green for available, red for occupied) and a dynamic counter to track available parking spots, ensuring high detection accuracy and real-time performance using Matplotlib.

• Implemented Edge Detection, feature matching Semantic segmentation, Image Processing, and Image clustering to analyze video feeds on large-scale image data and identify parking spots with 95% accuracy.

TWO-WHEELED ROBOT STABILITY                                                        

• Developed and simulated a self-balancing two-wheeled robot in MATLAB Simulink, using both PID and LQR control algorithms. Utilized IMU sensors to emulate real-world conditions and access them in GitHub

UNDERACTUATED ROBOT FINGER WITH 2-FOUR-BAR LINKAGE MECHANISM

• Designed a robotic finger with two four-bar linkage mechanisms, achieving human-like motion with a single actuator. 

• Analyzed degrees of freedom, forward kinematics, and singularities using SolidWorks for optimized performance and is available in GitHub.

RR ROBOT WITH CIRCULATORY TRAJECTORY

• Developed and implemented an iterative numerical method to solve the inverse dynamics problem for a rotating-revolute (RR) robotic arm. 

• Created a comprehensive mechanical model in Simulink and validated joint torque trajectories through forward kinematics and analytical simulations. Analyzed end-effector movements and confirmed accuracy with MATLAB plots and visualization tools and accessed them on GitHub

Robotic Arm using Arduino UNO embedded with sensors

• The robotic arm was developed by using Arduino UNO with human arm control using several sensors to control automatically.

Home Automation with Arduino Uno

• With the help of Arduino UNO and a Wi-Fi module, one can operate household stuff such as lights, fans, TVs, etc., via mobile remotely.

Calculator Application with Python

• Using Kivy in Python, A Calculator app was developed using Visual Studio code.

Smart bottle

• Smart bottle for heating/cooling liquid with a mobile app control wirelessly.

Smart Furniture

• Smart Furniture with smartest features such as wireless charger, smart kettle, smart speaker, etc.