A cloud-based Hospital Management System (HMS) offering Software as a Service (SaaS). The primary objective was to deliver a highly configurable HMS that operates exclusively on the cloud and is accessed through a single-page web application. All computational processes occur in the cloud, while the user interface remains localized. As a backend developer, my responsibilities included designing web services to expose data to the frontend of the application. I also conducted research and development to ensure the application's compliance with Health Level Seven (HL7) standards, successfully creating several proofs of concept. Moreover, I facilitated end-to-end integrations between HealthCloud and other established systems, such as Dynamics 365, to enable seamless synchronization of patient details between the Health Management Information System and the Enterprise Resource Planning (ERP) solution. In addition, I enhanced the application's code and applied various strategies to optimize its performance, particularly during high usage scenarios. I further introduced microservices to extract specific functionalities from the monolithic application. As part of a proof of concept, I explored the implementation of Cosmos DB to efficiently retrieve frequently accessed data, surpassing the performance of traditional relational databases.

I have expertise in profile writing and content creation, specifically helping developers enhance their profiles. Revised sentence: "A software solution designed for seamless integration, automating the collection of vital readings from patients to alleviate nurse workloads. It features a web-based portal for efficient management of reading rooms, which connect with a desktop-based reading location application that captures and processes readings from specific devices. Being a full-stack developer, I utilized ASP.NET MVC, Dapper, HTML, CSS, Bootstrap, JavaScript, C#, and jQuery to build the management portal. Additionally, I developed the reading room application using WPF, SQLite, and Material Design WPF, allowing for the retrieval of crucial patient readings from hardware devices. It also incorporates audio instructions through the Windows text-to-speech API for patients. Moreover, I implemented an Opentok WebRTC-powered nurse auto-ride feature, enabling seamless audio and video communication between doctors and the reading room. Furthermore, I created an HL7-compliant REST API server to receive patient information from any existing healthcare system, enhancing the application's versatility.

I have extensive experience as a full-stack developer, specializing in the integration module for the FICO LiquidCredit Small Business Scoring Service. My responsibilities included designing and implementing individual components for submitting FICO requests and processing credit scores for loan applications. Furthermore, I successfully developed credit memo customizations for specific lenders, utilizing Kendo jQuery UI components, and enhanced the existing UI controls. I also utilized Telerik Reporting to create essential system documents and reports modules for PPP loan forgiveness summaries and PPP credit recommendations. Additionally, my involvement extended to working on the Telerik RAD Controls-based PDF engine, ensuring the seamless mapping of SQL table data to and from the required fields of the PDF forms.

I was responsible for developing the backend of the biometric attendance management portal, which included building Windows services to operate on individual attendance terminals. These services handled biometric attendance devices, recorded and synchronized employees' time in and out, and ensured start and end times through thumbprint verification. Furthermore, I wrote REST APIs to process the attendance records from the terminals and manage them within the MS SQL Server database. In addition to my work on the biometric attendance system, I fulfilled the role of a full-stack developer by constructing the meeting room and workstation management module using ASP.NET MVC. This module allowed users to view, add, update, and remove rooms and workstations available within the office premises. Moreover, I designed and implemented the meeting room and workstation assignment module, catering to use cases involving booking meeting rooms, managing attendees, authorizing the start or cancellation of meetings, viewing assigned schedules, notifying start and end times, managing workstations or hot-desking, and notifying workstation assignments.

I designed and implemented a system that enables seamless processing of services such as mobile phone top-ups and bill payments through a dedicated terminal, similar to that of vending machines. In this role, I developed the back end of the self-service terminal server by crafting REST APIs utilizing ASP.NET Web API. These APIs facilitated various functionalities including CRUD operations, authentication, authorization, and terminal synchronization. Furthermore, I took charge of building the entire terminal application stack using WPF, SQLite, Dapper, and gridExtra. I successfully integrated multiple hardware devices such as a payout device, cash validator, card reader, and printer by designing suitable interfaces. Additionally, I researched and developed each involved hardware device, ensuring a well-rounded and efficient application. To provide uninterrupted functionality even in the absence of network connectivity, I implemented a transaction sync manager. Moreover, I constructed an Android application utilizing Xamrin forms to facilitate agent verification and terminal maintenance management. To enhance user interactions on the terminal machine, I created a WPF application, while also developing a web portal to manage administrative actions.

I have developed a cloud-based application and subsystem of NOVATRAQ that provides borrowers and referral sources with a secure environment to facilitate all necessary activities during the lending process for SBA. This includes offering customizable SBA loan application wizards, monitoring loan applications, and managing documents. In addition, I have created loan application interview wizards for SBA, SBA PPP, and SBA PPP forgiveness. I have utilized Kendo jQuery UI, Backbone.js, and Marionette to develop user controls and employed ASP.NET Web API with Entity Framework as the object-relational mapper to establish endpoint APIs. Furthermore, I have contributed to the client-side management aspect by implementing SignalR for monitoring user activity. Additionally, I have designed the product match and self-registration modules, enabling borrowers to initiate SBA PPP and its associated forgiveness loan applications. Lastly, I have developed the templates module with the use of Rackspace.net, granting borrowers access to template documents stored at Rackspace.

The application is specifically designed for Pakistani products and offers a functionality to suggest alternative medications based on the provided brand or generic name of a medicine. It allows users to easily find nearby pharmacies that have the desired medication in stock. Additionally, the app provides a specialized interface for pharmaceutical companies and pharmacies, enabling them to effectively manage their products, stocks, and branches. As a proficient full-stack developer, I skillfully developed separate ASP.NET MVC and Web API servers, utilizing jQuery and DataTables as my primary front-end development tools. Implementing custom, claim-based authorization and token-based authentication, I ensured secure communication between the Android application and Web API. Furthermore, I successfully utilized the Google Maps API to create reverse lookup queries for searching by both brand and generic names, as well as for the pharmacy locator module. Dedicated to expanding the application's functionality, I actively worked on various data scrapers, employing Selenium for .NET and different data cleaning and formatting techniques. Through this process, I was able to create a comprehensive dataset of alternative medicines specific to Pakistani products. Finally, I designed and implemented the manufacturer and pharmacy management modules using ASP.NET MVC.

I am the project's creator, responsible for developing a comprehensive solution that enriches the Kendo jQuery Grid by integrating server-side filtering, sorting, and aggregation features. The package extends support to .NET, .NET Core, and .NET Standard technologies.

I spearheaded the creation of a sophisticated Deep Learning-powered image classifier capable of accurately identifying over 200 distinct bird species. As a machine learning engineer, I meticulously constructed diverse models employing Convolutional Neural Networks (CNNs) and expertly fine-tuned each one to yield optimal outcomes. Throughout this process, I leveraged K-fold cross-validation techniques for enhanced model refinement. Notably, the most effective models harnessed the Inceptionv3 architecture, trained on the renowned ImageNet dataset. To further augment my dataset, I meticulously amalgamated multiple pre-existing labelled datasets, including the esteemed CUB and NABird datasets.

A comprehensive solution aimed at facilitating the recognition of printed Urdu text within images while accommodating various fonts. The conducted research not only produced an authoritative benchmark dataset for printed Urdu that incorporates diverse fonts, but also constructed sophisticated Deep Neural network models capable of accurately identifying Urdu text lines and words irrespective of the font utilized. Serving as the researcher, my role encompassed meticulously establishing a comprehensive rationale outlining the research's ability to surpass existing system limitations. Additionally, I meticulously collected, cleansed, and formatted the data, which consequently served as the foundation for generating the ultimate corpus. Furthermore, I leveraged cutting-edge CNN and RNN-based DNN models to successfully construct a fully-functioning OCR system. Lastly, my involvement extended to conducting an in-depth analysis of the dataset and the trained models.

I spearheaded the development of an advanced artificial intelligence (AI)-driven resume parsing system that enables candidate profiling and generates suitability scores based on the provided job description. This robust solution delivers a comprehensive end-to-end recruitment workflow, facilitating seamless job posting and efficient candidate shortlisting. As the team leader, I directed the creation of AI-assisted parsing, profiling, and scoring APIs. I meticulously crafted the code architecture of the Parsing API, adhering to industry best practices. Employing a test-driven development approach, I utilized Python's unit test library and ensured compliance with PEP standards through the implementation of Flake8. My responsibilities further encompassed the design of various modules within the application, including PDF parsing, resume structure extraction (columns, headings, and sections), resume sort order correction, template-based parsing, and scoring and rating functionality. NLP-based techniques, particularly the utilization of a fine-tuned version of Spacy NER, were instrumental in extracting specific keywords. Additionally, I led the research and development team in preparing training data for fine-tuning the Spacy NER (BERT) model. Furthermore, I guided the team in constructing an ML model that leveraged styling and formatting information for the classification of resumes, utilizing style occurrences as relevant model features.

This advanced system employs cutting-edge deep learning models to facilitate precise face detection and recognition, enabling the seamless tracking of patients' activities within a medical facility. It effectively monitors both their entry and exit through video streams, while also serving the purpose of accurately timing therapy sessions. In my role as the research and development lead, I successfully constructed robust face detection and recognition systems utilizing the superior capabilities of ArcFace and Facenet512. This involved creating an ensemble model for reliable face recognition, as well as employing SSD for dynamic face detection in real-time video streams. Additionally, I provided invaluable assistance to the development team in designing the application architecture. This encompassed establishing a streamlined patient registration process using face recognition and devising an edge-based detection and recognition system to enhance communication efficiency with the existing monolithic EHR system. Moreover, I oversaw the development of a FastAPI-based monolithic server that proficiently managed various endpoints for recognition purposes.

I have expertise in creating high-quality content and crafting compelling profiles. I collaborate with developers to enhance their profiles. Allow me to rephrase the given sentence to sound more professional: "Our platform is designed for the development and deployment of sophisticated bots that embody the distinctive qualities of renowned personalities, enabling them to perform various tasks. These bots possess the capability to handle email and SMS communication, respond to tweets, endorse products, and even engage in marketing activities on behalf of the personality they represent. Moreover, the bot offers advanced functionalities, such as generating voice messages that perfectly mimic the personality's tone and voice. In this role, I collaborated closely with a team of skilled machine learning (ML) engineers to thoroughly explore extensive language models (LLMs) and devise techniques to restrict their knowledge and style to a specific celebrity. I conducted comprehensive research on diverse embedding models to extract relevant information that matches input queries, which served as valuable data for better instruction and fine-tuning of the LLMs. Additionally, I focused on optimizing the process by implementing solutions like leveraging Milvus for accelerated vector searching and exploring cost-effective, open-source LLM options such as Llama-V2-7B-chat. To provide a seamless solution, I developed a comprehensive backend system based on FastAPI, enabling the exposure of our solution. I also architected a multi-tenancy infrastructure with configurable workflows tailored to accommodate multiple personalities. Furthermore, I successfully integrated text-to-speech and speech-to-text modules using Whisper and ElevenLabs, respectively, enriching our offering with audio input and output functionalities.

I have collaborated with the development team to create a Visual Studio code extension that utilizes our internally hosted LLM to generate code completions. This ensures that our proprietary client code remains confidential and is not exposed to any external service. Throughout the project, I served as a research consultant, working closely with the team to shape the idea of enhancing client code security within our organization. We extensively investigated various existing tools and LLMs, conducting a thorough comparative analysis of their performance and efficacy in relation to our specific use case. After careful evaluation, we concluded by modifying the existing VS Code extension to align perfectly with our requirements, enabling it to be easily configured to integrate with our locally hosted LLM. Lastly, we successfully deployed the Code Llama-13b on RunPod using the Hugging Face Text Generation Inference (TGI) library, granting API access to the development team.

Developed a cutting-edge ingredient extraction system to retrieve data from restaurant menus. The data was subsequently utilized to establish a connection between retailers and restaurants by identifying the ingredients sold and consumed, effectively pinpointing targeted restaurants for retailers. Constructed a comprehensive data preprocessing pipeline to implement specialized techniques for cleansing and normalizing the information. Additionally, devised a two-step ingredients extractor that initially attempted to extract ingredients from the provided menu data and then prompted the Language Learning Model (LLM) to fill in any missing information during the second pass. Currently, this project is actively being developed, incorporating an in-house machine learning model to eradicate the need for LLMs and create a highly scalable solution.

I developed a machine learning system to identify duplicate products from various food delivery platforms like UberEATS, Zomato, HungerStation, Mrsool, and Jahez. This system enabled price comparison and identification of missing products across different platforms. Additionally, I served as the lead data scientist, responsible for identifying and gathering relevant data to create valuable features. Moreover, I designed a system to extract product attributes from the available data and utilized these features to construct a semi-supervised tagged dataset comprising 50,000 examples. Subsequently, multiple machine learning models were trained on this dataset to determine similarity scores. Our system not only facilitated the identification of common stores and comparison of their products on different food delivery platforms but also effectively eliminated duplicate store entries extracted by our crawlers. This proposed system exhibited an outstanding accuracy rate of 98% and was successfully utilized to process over 3 million product records.

I have developed an add-on for Reki, a cutting-edge platform designed to effortlessly track recommendations from friends and facilitate discussions about watched shows. Through the utilization of LLMs, this add-on tackles the challenge of cold start problems by providing personalized movie recommendations based on an initial questionnaire. Additionally, I successfully constructed an end-to-end solution using OpenAI GPT-4 to generate a TMDB filter for extracting comprehensive movie information. Making use of specific NLP techniques, I effectively parsed the irregular format of the LLM output. Furthermore, I conducted experiments with Mistral and Llama models to explore alternatives to OpenAI GPT-4 in order to mitigate scaling costs. Lastly, I implemented a secure API key-based authentication system to facilitate secure communication between the Reki app and our recommender.

Developed a sophisticated machine learning and natural language processing solution to extract retail products and associated attributes for comprehensive product sales analysis and competitive market assessment on platforms such as UberEATS, Zomato, Hunger-Station, and Jahez. Served as the lead for machine learning implementation, leveraging cutting-edge techniques to construct an advanced ETL pipeline capable of identifying and extracting brands, quantities, and other relevant attributes. Devised a streamlined data cleansing and normalization pipeline utilizing NLP methods such as tokenization, expression evaluation, stemming, and translation for unlabeled data inputs. Enhanced system performance through the integration of Dask data frames as a replacement for pandas, significantly improving processing time. Implemented dedicated modules for confidence calculation, post-processing, and verification based on extensive domain expertise gained from client interactions. Successfully tagged over 10 million records with the necessary information using the final tagging system.

I spearheaded the development of a machine learning (ML) system designed to compare and identify common stores by analyzing store data from various food delivery platforms. This duplicate identification process served as the foundation for data analysis and correction of existing tagged data. Additionally, my role as the lead data scientist involved identifying and curating relevant data to generate valuable features. I also designed and implemented a conventional Natural Language Processing (NLP)-based system, which involved human collaboration to tag approximately 40,000 examples. Moreover, I conducted extensive experiments using various ML and DL models, specifically fully connected networks, in order to predict similarity scores. While Siamese networks were experimented with to explore similarities, their efficacy did not meet expectations. Furthermore, I successfully implemented geo-fencing techniques and utilized geo-hashing libraries to efficiently group stores and minimize matching confusion. This resulted in improved overall system efficiency. The same system was utilized to identify and eliminate duplicate store entries extracted by our crawlers. The proposed system achieved an impressive accuracy rate of 96% and effectively processed over 500,000 records.

A tool that analyzes calls between customer service e agents and clients, scoring the user experience and call quality using state-of-the-art Artificial Intelligence. It helps organizations manage and track the satisfaction of their customers and manage the overall quality of support and marketing calls. I researched and developed a complete ML pipeline from processing and transcribing audio calls, to speaker diarization, and generating a score based on configurable domain-related questions for scoring a call between a client and a customer service person using LLMs. I crafted a FastAPI-based product around the ML solution created. I architected the application adding authentication, multi-tenancy support, and creating endpoints for exposing data for analytics. I created a module for providing a queue-based mechanism for processing call-analysis requests. I built an integration with Stripe for managing user payments.

Outset style provides a platform for managing a digital closet. It provides features to track purchase details, analyze shopping insights, and view and organize closet listings and outfits. I worked as a full-stack developer creating features and fixing the issues with the existing version of the application. I architected the new backend architecture improving the existing code base. I developed multiple modules including token-based authorization, syncing products from Gmail in real time using Gmail API. I created an LLM-based email parser as a fallback strategy for extracting clothing article details from emails if the email cannot be parsed by the existing rules-based system.