Our Portfolio

This project develops a web-based application utilizing Python 🐍 and Django ⚙️ to facilitate collaboration among researchers by indexing and searching scholarly articles. The system leverages Google Scholar's 📊 API to retrieve relevant data, providing personalized search results based on user preferences. Additionally, the application incorporates natural language processing techniques using NLTK 📝 to extract relevant information from article abstracts and keywords.

We developed an Attention-based LSTM algorithm 🐍 for generating captions in Urdu for provided images. This project involved creating a dataset of images with Urdu captions and training the model for caption generation, resulting in a successful online implementation.

We propose and evaluate a deep reinforcement learning (DRL 🤖)-based algorithm for PHY layer transmission parameter assignment in LoRaWAN, utilizing LoRa as its physical layer. The existing rule-based algorithms contribute to packet collisions, degrading the performance of densely-deployed low-power wide-area networks (LPWANs). Our DRL-based algorithm ensures fewer collisions and better network performance compared to state-of-the-art algorithms for LoRaWAN, achieving up to 500% improvement in PDR in some cases. This research was published in IEEE Local Computer Networks (LCN) 2020 https://scholar.google.com/citations?view_op=view_citation&hl=en&user=YHSAFvEAAAAJ&citation_for_view=YHSAFvEAAAAJ:UeHWp8X0CEIC. A continuation article is available at [https://scholar.google.com/citations?view_op=view_citation&hl=en&user=YHSAFvEAAAAJ&citation_for_view=YHSAFvEAAAAJ:2osOgNQ5qMEC](https://scholar.google.com/citations?view_op=view_citation&hl=en&user=YHSAFvEAAAAJ&citation_for_view=YHSAFvEAAAAJ:2osOgNQ5qMEC].

Recent advances in deep learning have sparked significant excitement around harnessing rich information hidden in large data volumes and tackling complex problems. This enthusiasm has led to tremendous interest in applying artificial intelligence/machine learning (AI/ML) based network automation, control, and analytics in 5G and beyond. This article presents a cautionary perspective on AI/ML use in the 5G context by highlighting the adversarial dimension across various machine learning types, including supervised 📊, unsupervised 🤖, and reinforcement learning 🏋️‍♀️. The article supports this through three case studies and discusses approaches to mitigate this adversarial ML risk. Additionally, it offers guidelines for evaluating the robustness of ML models and highlights issues surrounding ML-oriented research in 5G more generally. This research has been published in a top conference (IEEE Internet Computing 2021): https://ieeexplore.ieee.org/abstract/document/9314254/.

Developed a DQN (Python 🐍) and DDQN (Python 🐍)-based Deep Reinforcement Learning algorithm to play Flappy Birds and Lunar Lander games. The algorithms successfully played those games for an extended period, achieving high scores.

We developed computer vision-based algorithms utilizing Convolutional Neural Networks (CNN) 📊 to classify objects and numbers into predefined categories and sub-categories. Furthermore, we refined these trained models for additional object classification, yielding notable outcomes.

Developed a semantic matching algorithm using word embeddings for a large corpus of NEWS articles 📰. The corpus was gathered from numerous NEWS websites over the internet utilizing specially designed scrapers 🕸️. The matched articles were populated in a graph database called Neo4j 📈. An API was developed to display relevant articles to the article being viewed by a client, presented as a comparison graph 🔢.

Developed a K-Means-based algorithm 🐝 in Python for efficiently assigning pre-defined delivery orders from various vendors to a fleet of riders, incorporating numerous business logics into the process. Leveraging this algorithm and methodology enabled the client to significantly reduce assignment time, previously handled by a team of professionals. As a result, daily order assignment and acceptance counts increased from 1000 orders per day to approximately 4000 orders per day.

Developed a Graph Convolution Network (GCN) 📈-based algorithm for fraud detection, utilizing an online dataset. The dataset was divided into 80% training set and 20% testing set. Initial results were considerable. To enhance the model's robustness, adversarial samples 💣 were used to attack the models, observing a reduced performance. By fine-tuning the model on these adversarial samples, robustness against similar attacks was achieved, ultimately enhancing overall performance.

We developed and tested various configurations of Convolutional Neural Networks (CNNs 🤖) for detecting COVID-19 from X-ray images. A binary dataset was utilized, comprising 2 classes: "Infected" and "Normal". The infected class includes chest X-rays of individuals with Pneumonia and Covid-19 infections, while the normal class consists of normal chest X-ray images. Research findings are publicly available at https://github.com/Inaam1995/MSCS18037_COVID19_DLSpring2020?tab=readme-overview.

I developed a note-taking application that offers advanced features similar to those found in Notion, including "Chat to Document" functionality 📝, automated Python 🐍 "Document Summarization" capabilities 🔍, real-time machine learning-based "Document Translation" 🗣️, live collaboration tools 💻, and comprehensive user authentication procedures 👥.

Developed a Python 🐍 AI-powered virtual sales advisor that utilizes scraped data from multiple car-selling platforms and integrates an intelligent recommendation engine to assist users in finding the most suitable car based on their budget, preferences, and usage needs. The system effectively processes natural language queries, filters results by multiple parameters (price, brand, mileage, condition), and provides ranked recommendations with alternatives when exact matches are unavailable. It also supports English 🇬🇧 and Arabic ⏰ scripts queries, and is designed to scale for multiple car marketplaces. Outcome: A scalable recommendation agent that improves car discovery by turning user preferences into actionable, ranked suggestions.

Utilizing Microsoft Word 📄, a series of LinkedIn posts was developed to enhance the client's personal brand, establish thought leadership, and attract potential leads. The posts were grounded in factual, verifiable information and showcased the client's distinctive writing style across economics, data analytics, leadership, growth, and sustainable development. Outcome: Ready-to-publish, brand-consistent content aligned with the client’s expertise and optimized for engagement and lead generation.

Developed a system that interprets medical questions and provides accurate procedure suggestions utilizing a Retrieval-Augmented Generation (RAG) pipeline. The solution integrates trusted medical sources with Large Language Model (LLM 🤖) technology to generate evidence-based recommendations supported by citations. Outcome: A clinical support tool that provides ranked procedure suggestions with evidence citations and contraindication alerts. (Note: For decision support only. Requires healthcare professional verification.)

I designed and implemented a comprehensive backend system for REST APIs within an Order Management System, utilizing MongoDB ⏹️ as the primary database. The system was deployed on Heroku 🚀. The REST API suite includes: * Order Creation API * Order Editing API * Order Update Manager API * Single Order Display API * Multi-Order Display API with advanced filtering capabilities * Order Assignment to Fleet API * Eligible Rider Finding API * Order Status Management APIs

As a technical lead, I developed a comprehensive backend system for a Fleet Management System's REST APIs. This endeavor included designing and implementing REST APIs for a mobile application 📱, driver portal 📊, onboarding panel ✈️, and driver management portal 👥. The schema was designed for required databases and tables 📈, utilizing MongoDB 🐒 as the primary database. The backend was deployed on Heroku 🔩. Key APIs developed include: * Order Status Update API * Single Order Display API * Multi-Order Display API * Order Assignment to the Fleet API * Driver Onboarding/SignUp APIs * Driver Management APIs (Update Rider Details) * Driver Blocking/Unblocking/Unavailability APIs * Location Tracking APIs

I developed a comprehensive backend system for REST APIs supporting an AI assistant mobile app, utilizing AWS DynamoDB 🚀 and deploying on AWS Lambda ⏰. The system includes designing database schemas and tables. The REST API suite comprises: * User SignUP API * User SignIn API * User Conversation API * User Forgot Password API * Get all conversations API * Get the latest active conversation API * Upload Image API 🔗 * Upload Audio API 🎵 * Upload Video API 📹 * Live Video Socket Implementation ⏺

I developed a custom script using Python 🐍 to retrieve hotel pricing data from Marriott's website (www.marriott.com 📊) at specified locations and timings. The script generated a table of prices for hotels near the specified location and formatted it in HTML format before sending it to the client via email. To populate the specified locations, I utilized a MongoDB database 💻 with a custom frontend built specifically for this purpose. The script involved performing specific steps to reach the required data, including updating filters as needed during the process.

This project entailed developing a custom script using Python 🐍 to periodically gather and update information on the top 100 USDT trading pairs, along with their corresponding pricing data, from Binance's main website 🏗️. The extracted list was then utilized by a bot that regularly monitored price fluctuations in these trading pairs.

Here is the rewritten project description: This project involved designing and implementing a custom script to automate document updates on a website 🌐. The process entailed searching for a company, navigating to its respective page, accessing the document directory, and downloading each PDF file 🔴 individually. Following download, the files were hashed and verified for duplicates. To evade detection by the website's bot mechanism, NordVPN 🛡 was utilized to regularly update our bot's location.

A Python 🐍-based automated script was designed and developed to scrape specific product categories from Amazon's website, collecting relevant information such as names, descriptions, product IDs, prices, and more. The pricing of these products was then compared to that provided by the client for similar products on other websites. A comprehensive spreadsheet is generated, detailing the different prices of the same product across Amazon and other websites. The script runs every four hours, producing a new spreadsheet that is transmitted to the client via email or file-sharing platforms.

A Python 🐍 script was developed to automate data collection from specific product categories on eBay's website, extracting essential details such as names, descriptions, product IDs, prices, and more. The collected pricing data was then compared with that of products on other websites provided by the client. A comprehensive spreadsheet was generated, featuring the different prices of the same product across eBay and other platforms. The script runs every four hours, producing a new sheet and transmitting it to the client via email or file-sharing tools.

This project entailed developing an automated script in Python 🐍 that scraped wine pricing information from the wine-seeker's website. The script collected names, prices, and providers of various wines, then compared their prices across different providers on the same website. A spreadsheet was generated to display which wine is offered by which provider at the cheapest price. The script runs every four hours, creating a new sheet and sending it to the client via email or file-sharing platforms 📨. To circumvent the website's bot detection mechanism, trial-and-error techniques were employed to simulate human-like behavior, thereby evading detection.

This project entailed developing an automated script that efficiently scraped TripAdvisor data using Python 🐍. The extracted textual information was stored in a PostgreSQL database 📊, while categorized location data was organized into separate databases. Images were uploaded to Dropbox 💾. The script operates on a regular schedule, continuously updating the dataset with new locations and images. To facilitate efficient scraping of large datasets, rotating proxies ⚡️ were utilized to enable parallel data collection.

Using Python 🐍, an automated script was developed to scrape relevant data from masstimes.org. Extracted information about churches, including mass timings, was formatted into JSON files. To facilitate data extraction, US Zip codes were utilized to simulate searches on the website, as no central search methodology was provided. Parallelization techniques were employed to efficiently process the large count of churches and maximize speed. The extracted JSON files were then saved on Google Drive 📁. Rotating proxies were used to enable parallel data extraction due to the large dataset size.

A Python script was developed to scrape data from directory.charlestondiocese.org, extracting comprehensive information about churches, including mass timings and assignment details, into JSON files. To optimize processing efficiency for a large number of churches, parallelization techniques were employed. The extracted JSON files were saved on Google Drive 📁, a file-sharing platform. Furthermore, images were extracted from respective churches and stored according to the client-provided naming convention.

I collaborated with Spoonful Inc. to develop a large-scale data extraction pipeline 🚀 using Python 🐍 for major grocery store websites, enabling the client to centralize product data for market analysis and inventory management. The objective was to accurately extract detailed product information, including ingredients 👩‍🍳, allergens ⚠️, nutrition facts 🥑, and product metadata 💻, while ensuring compliance with each site's access structure 🔒.

Developed a Python-based trading bot for CSFloat that monitors listings, identifies underpriced skins using configurable thresholds, executes automated purchases, and relists items for profit. Integrated smart analysis for float value, rarity, and pricing, providing a fully automated, efficient trading workflow. Scope of Work: Deal Monitoring & Analysis: Monitors active CSFloat listings to detect underpriced skins by comparing each listing against similar items (considering wear, float value, rare patterns, and stickers). Profitability Logic: Applies configurable thresholds, e.g., only purchasing items at least 10% cheaper than the next lowest comparable listing. Automated Trading: Automatically purchases selected skins and relists them on CSFloat at competitive, profitable prices once trade hold periods expire. Custom Configuration: Supports user-defined parameters for price thresholds, specific skin types, and trading rules. Scalability: Designed to integrate additional trading logic or marketplaces in the future.

Utilizing Python 🐍, an automated job application system was developed to streamline the job application process across multiple platforms, including Workday 💼, Greenhouse 🌿, ATS Ripple ⚖️, and ASBQ Jobs 📊. The system automates the filling out of application forms, uploading of user resumes, and submission of applications with minimal human intervention.

Developed an automated system that extracts and analyzes court case records from U.S. county websites, identifying commercial contract cases and generating structured Excel reports 📊. The system leverages APIs, including AccurateAppend 📲 and Enformion 🔒, to enhance data with verified mobile numbers and emails for compliant SMS and email campaigns 📨.

A technical solution was developed using Python 🐍 to automate the identification and processing of Merchant Cash Advance (MCA) court cases from public databases. The system efficiently filters relevant cases, enhances them with verified business contact information, and delivers daily lead reports via email.

A Python 🐍 automation bot was developed to streamline lead extraction from the client's CRM, directly uploading them into Kixie 📊. The system efficiently organized leads by type and zip code, generated separate CSV files ⚙️, and ensured seamless syncing between platforms 🔄. Furthermore, an additional module was integrated to extract verified phone numbers from PlanetAltig 💻 website while avoiding duplicates or previously processed leads 🔒.

A real-time stock data automation system was developed for StockTitan.net 📊, utilizing Python 🐍 scripts to extract, process, and display live market data updates every few seconds. The system fetched live data fields including stock symbol, volume, price, time, % change, and float. A custom-built web app interface powered by HTML 💻, CSS 💻, and JavaScript 💻 displayed the top 10 momentum stocks. Additionally, a voice automation module was implemented to audibly announce the latest stock symbol in real-time, enabling seamless live YouTube broadcasts.

Python 🐍 scripts were developed to automate data extraction from a client's website using APIs 📊. The scripts accept betting codes as input and produce structured data, ensuring accurate and reliable results. Furthermore, three key APIs were extracted from the site, and existing code bugs were corrected for seamless operation 💻.

Developed and optimized database queries using MongoDB 🐍 and Neo4j 🔥 for a client's assignment exercises. The project scope included editing and implementing 14 exercises, automating data imports, and verifying accurate query results. Additionally, comprehensive written documentation was prepared to accompany the queries.

The project entailed automating data retrieval from client-shared websites utilizing backend APIs 🐍. Through monitoring website traffic, identifying pertinent API endpoints, and crafting custom Python 🐍 scripts, I successfully fetched, cleaned, and processed the extracted data.

I developed a simple and functional frontend utilizing HTML 📋, CSS 💅, and JavaScript 🐍 to integrate with a backend Python 🐍 script created in a previous order. The interface enables users to easily add and remove product IDs. Furthermore, I constructed a custom Amazon scraper to automatically extract product data as requested.