Things I’ve made trying to put my dent in the universe.

I’ve worked on tons of little projects over the years but these are the ones that I’m most proud of. Many of them are open-source, so if you see something that piques your interest, check out the code and contribute if you have ideas for how it can be improved.

    leave blank

    Calling Agents Demos

    Developed inbound and outbound multilingual calling bots to enhance customer interaction and sales efficiency. Achieved a 6% increase in sales and 90% accuracy in conversational tasks.

    demo calls

    leave blank

    CFO-II

    Created a voice-to-voice financial analysis agent using Retrieval-Augmented Generation (RAG) and Function Calling.

    hosted demo

    leave blank

    Pre-training BERT and GPT2 Family Models for Papia Language

    Built a pipeline for training tokenizers and language models tailored to the Papia language.

    github repo

    leave blank

    OB/GYN Exam Preparation Bot

    Developed a chatbot to assist users in preparing for exams and interviews in the OB/GYN field.

    github repo

    leave blank

    Website Analyzer

    Created a website content analysis tool that summarizes and critiques site content dynamically.

    github repo

    leave blank

    Real-Time Whisper Client Server

    Developed a real-time transcription system using Whisper ASR for highly accurate and low-latency speech-to-text conversion.

    github repo

    leave blank

    Alee TTS: Conversational Text-to-Speech System

    Developed an advanced Text-to-Speech (TTS) system capable of producing natural and conversational speech, including nuanced expressions like laughter, 'hmm,' and 'umm.'

    Alee TTS Demo Audios

    leave blank

    Modified DINO with Masked Image Modeling

    Enhanced Vision Transformer (ViT-Small) by integrating DINO and Masked Image Modeling (MIM) techniques for pretraining on small datasets. Improved model robustness and performance in self-supervised learning tasks.

    github

    leave blank

    PillWise

    Developed a Multimodal Retrieval-Augmented Generation (RAG) agent for medication identification. Paired with a mobile app, the system detects and identifies medications through images and allows conversational interactions about the medication.

    github

    leave blank

    Image Matting for Background Removal

    Developed a model for accurate image matting to enable seamless background removal and compositing for e-commerce and media applications.

    github

    leave blank

    AI-Based Histopathological Diagnosis

    Built convolutional neural network (CNN) models for diagnosing histopathological images of lung adenocarcinoma and squamous cell carcinoma.

    github

    leave blank

    COVID-19 Diagnosis Using Swin Transformer

    Fine-tuned Swin Transformer for diagnosing COVID-19 from chest X-ray images, outperforming existing CNN-based approaches.

    github

    leave blank

    Pneumonia Detection from Chest X-rays

    Designed a hybrid deep learning model combining CNNs and tree-based algorithms for pediatric pneumonia detection.

    github

    leave blank

    Rust Detection Using YOLOv8

    Created an object detection model for identifying rust on machinery and vehicles to aid in predictive maintenance.

    leave blank

    Facial Search System

    Developed a FaceNet-based system for embedding and retrieving facial data with high accuracy.

    github

    leave blank

    Sign Language Recognition

    Final Year Project (FYP) focused on recognizing sign language gestures using Shape Moments Analysis.

    leave blank

    Object Detection in Medical Imaging

    Developed a detection system for identifying anomalies in medical imaging datasets.

    leave blank

    Virtual Reality Racing Game

    Mentored students on developing a Unity 3D-based VR racing game with real-world physics and enhanced graphics.

    leave blank

    Histopathological Image-Based Question Answering

    Fine-tuned multimodal Microsoft's Florence-2-large model to enable question-answering on histopathological images for medical diagnosis and research.

    huggingface repo

    leave blank

    Text-Based Audio Cropper

    Designed a system to transcribe and analyze audio segments using speaker diarization and word-level timestamps.

    github repo

    leave blank

    Search Engine for Syntax and Contextual Similarity

    Designed a hybrid search engine combining syntax-based fuzzy matching and Sentence-BERT embeddings.

    github repo

    leave blank

    Sales Call Analysis

    Built an NLP pipeline for analyzing sales call conversations to detect key insights such as pricing issues, tone discrepancies, and salesman negligence.

    leave blank

    Text Summary Generator

    Fine-tuned T5 models to generate concise and contextual summaries from long-form research papers.

    github repo

    leave blank

    Ecommerce Sale Agent Bot

    Built a chatbot capable of handling FAQs across multiple languages with dynamic response generation, sale agent, recommend products of Ecommerce website.

    hosted demo

    leave blank

    Pretraining BERT AND GPT for Papialanguge

    Developed pre-trained language models specifically for the low-resource Papia language, with applications in text classification, sentiment analysis, and translation tasks.

    github repo

    leave blank

    Real-Time ECG Monitoring

    Developed a real-time system to monitor ECG signals and provide actionable insights for healthcare applications.

    leave blank

    1D ResNet Models for ECG Signal Classification

    Built and deployed 1D ResNet models to classify ECG signals and detect abnormalities with high accuracy.

    leave blank

    AI-Driven Trading Indicators

    Designed and implemented AI-driven trading indicators to predict profitable trades based on historical time-series data.

    github repo readme

    leave blank

    Chart-Based Image Classification for Trade Predictions

    Developed and fine-tuned models to classify trading charts and predict trade directions and types.

    github repo

    leave blank

    ECG Signal Conversion from Images to Time Series

    Designed a technique to convert ECG images into structured time-series signals for downstream analysis.

    leave blank

    Anomaly Detection in Time-Series Data

    Created anomaly detection models to identify irregularities in industrial machine operations.