My Portfolio
Tech Stack: HTML5, CSS3, JavaScript, Full Stack Web Application, GitHub Pages
- Expertise in media queries and mobile-first design ensuring an optimal viewing experience on various devices.
- UI/UX Enhancements smooth scrolling, a typewriter effect, animated gradients, and hover effects that enhance user engagement and navigation experience.
- Demonstrates precision in designing interactive elements and ensuring both aesthetic appeal and functionality in the user interface.
- Use of SEO-friendly meta tags and performance strategies (e.g., suggestions for gzip/Brotli compression) to improve site visibility and load times.
SmartPay UPI – QR based Payment System
Tech Stack: Blockchain, Python, OpenCV, QR Code APIs, SMTP, bcrypt, JSON, CSV, Privacy |
GitHub Repo
- Unique QR Based Payments linked to user accounts for hassle-free payments through a simple scan.
- Seamless and secure financial transactions tailored to the Canadian market, integrating QR code functionality for instant payments.
- Real-time processing with blockchain-based transaction integrity, role-based access control, and bcrypt-enabled password hashing.
- Admin dashboard for managing user data, monitoring blockchain validity, and detecting tampered transactions.
PropertyInsight: Real Estate E-Commerce (2024)
Tech Stack: Java, Selenium, OpenCSV, Apache Commons CSV, Data Structures (AVL Trees, Tries), Boyer–Moore String Search, Regex, Multi-threading, Logger APIs |
GitHub Repo
- Efficient Search with Autocomplete: AVL Trees provide real-time autocomplete suggestions for city names, ranked by frequency to enhance user experience.
- Real-Time Web Scraping: Collected property data from websites like Remax and Zolo using Selenium, automating the extraction of price, address, and details.
- Advanced Filtering and Ranking: Supports filtering properties by price, city, province, and bedrooms/bathrooms, with keyword-based property ranking using Boyer–Moore and inverted indexing.
- Spell-Checking and Data Cleaning: Trie-based spell checker ensures accurate city/province searches, while data cleaning normalizes inputs and integrates data from multiple sources.
Scalable Big Data Architecture with Zstandard Compression for IoT Smart City Environments (2024)
Tech Stack: Shell, Python, Apache NiFi, Kafka, Spark, HDFS, Zstandard, Docker, Kubernetes |
GitHub Repo
- Collaboration: Led a four-member team to design and implement distributed data pipelines for smart city IoT analytics.
- Efficient Data Processing: Utilized NiFi, Kafka, and Spark for seamless data streaming and applied Zstandard compression to improve processing efficiency by 30%.
- Scalability and Orchestration: Deployed architecture on Kubernetes and Docker for scalability and reliability in handling high IoT data volumes.
- Impact: Reduced end-to-end processing latency by 30% and improved resource utilization by 40%, enhancing overall system efficiency.
TermAI Infinity: Offline Advanced LLM Toolkit
Tech Stack: Python, Transformers, Hugging Face, LangChain, PyTorch, Chroma, Local LLMs |
GitHub Repo
- Advanced Local LLM Capabilities: Integrated modules for text generation, retrieval-augmented generation (RAG), multi-step reasoning, summarization, and iterative text refinement.
- Efficiency & Modularity: Built with modular components like Summarizer, Refiner, and MultiStepReasoner, allowing customizable workflows for offline environments.
- Retrieval-Augmented Generation: Implemented RAG pipeline using Chroma vector stores and Hugging Face embeddings to provide context-aware answers from local text files.
- Customizable CLI Tool: Provided a command-line interface supporting operations like text generation, file summarization, and chain-of-thought reasoning for diverse use cases.
EdgeAIOptimizer: High Performance ONNX Inference
Tech Stack: C++, Python, ONNX Runtime, OpenCV, PyTorch, Custom Optimization Algorithms |
GitHub Repo
- Advanced AI Inference Framework: Designed a robust C++ engine to execute ONNX-based AI models on edge devices, supporting rapid and accurate decision making in resource constrained environments.
- State-of-the-Art Model Optimization: Implemented quantization, operator fusion, and ONNX graph-level enhancements, achieving up to a 3x improvement in inference speed without compromising accuracy.
- Custom Preprocessing Pipeline: Built a tailored preprocessing module with OpenCV for seamless image resizing, normalization, and conversion to ONNX-compatible tensors, enabling efficient data ingestion.
- Collaborative Development: Successfully integrated modular components for preprocessing, and post-processing, ensuring extensibility for future edge AI use cases.
Efficient Image Restoration for Noisy and Low-Resolution Data
Tech Stack: Python, PyTorch, OpenCV, Skimage, PIL, NVIDIA CUDA, TorchQuantization |
GitHub Repo
- Innovative Model Architecture: Implemented a custom UNet-based architecture for image denoising and super-resolution tasks, optimizing computational efficiency and accuracy for diverse input data.
- Quantized Deployment: Optimized the model for edge deployment by leveraging PyTorch's dynamic quantization, reducing inference time and memory footprint while maintaining accuracy.
- End-to-End Pipeline: Developed a complete training and evaluation workflow, including dataset preparation, model training, real-time inference, and automated performance analysis for seamless experimentation.
- Scalable Design: Integrated modular dataset handling, allowing scalability to new datasets or additional image restoration tasks with minimal adjustments.