Suleman Hamdani Email & Phone Number

Islamabad, Pk

Islāmābād, Pakistan

• Maintained and improved a Retrieval-Augmented Generation model (SiteJetAI) by fixing a bug that limited its contextual capabilities. Applied techniques like query decomposition improving its capability to answer advanced queries.• Enhanced SiteJetAI performance and deployed using Docker, Postgres, Flask, ElasticSearch and OpenAI API decreasing the response time by 80%.• Migrated 40+ Django-based endpoints to FastAPI, seamlessly integrating the new FastAPI-based flow with both the existing and new front-end systems. Additionally, converted 15+ database tables from Django ORM to SQLModel, ensuring efficient and scalable data management. Replaced Django serializers with Pydantic models to leverage FastAPI’s native data validation and serialization, resulting in a more streamlined and performant codebase.• Developed and maintained a FastAPI-based backend supporting dynamic API routing, handling multi-model database interactions, and optimized for scalability and low-latency processing.• Designed and implemented RESTful APIs for MCQ creation and management, improving efficiency in educational content handling by 40%.• Engineered a custom XML parser for Microsoft Word documents using lxml, automating data extraction, reducing manual processing time.• Collaborated with the frontend team to develop an FastAPI backed in-Word add-on for automatic and AI-based grading, enhancing grading efficiency and accuracy. • Leveraged GPT-3.5 API calls along with complex prompt engineering and open-source frameworks like Deepeval to create a descriptive answer grader, providing scores on metrics such as relevancy and correctness for enhanced assessment accuracy.• Utilized ngrok for re-routing and testing during development, enabling seamless local testing and remote integration with external services.• Customized and integrated Open edX with a FastAPI backend and local MySQL database for enhanced functionality and API-driven operations.

• Used satellite imagery and machine learning algorithms to geo-locate air pollution assets in South Asia.• Developed an AWS-based pipeline from scratch to process and detect brick kilns in over a million Sentinel-2 tiles.• Utilized AWS EC2 instances, S3 buckets, and parallel processing to iterate through 1.2 million square kilometers of the IGP region, reducing processing time by 72-fold. • Maintained APAD’s AWS account, including requesting increased resources and writing detailed use cases to ensure the allocation of up to 72 vCPUs, which significantly enhanced processing capabilities.• Developed a pipeline using SentinelHub to automate the mosaicking and retrieval of Sentinel-2 imagery for the entire region of Bangladesh, removing any sort of manual labour which mosaicking requires• Used Google’s Static Maps API to build a dataset of brick kilns in Pakistan and India. Trained YOLOv8 models using the collected PNG images, followed by conversion to TIFF format for accurate geolocation. • Engineered a scalable pipeline to process over 200,000 satellite images using Google Static Maps API on AWS G4 instances.• Integrated API for image retrieval with a 100% success rate, efficiently handling high-volume data.• Utilized YOLO for kiln detection, achieving 95% mAP @ 0.5 and 60% mAP @ 0.95. • Calculated geographic coordinates from pixel data using linear interpolation, ensuring 100% spatial accuracy.• Transformed bounding box pixel coordinates to geographic latitude and longitude based on image resolution and map scale.• Detected and validated 55,000+ brick kilns in the IGP region using the above mentioned pipeline.

Islamabad, Pakistan

• Led a classification project employing multi-spectral Landsat8 Satellite Imagery to semantically segment wheat, rice/paddy, and soybean crops across 20+ cities, covering 12 million hectares of cultivated land. • Analyzed segmented crops’ health through vegetation indices, achieving 98% classification accuracy and 94% Kappa score.• Leveraged Google Earth Engine to annotate sentinel 2 satellite imagery from 10 plantation sites, covering 1 square kilometer area and 4 unique classes with an accuracy of 93%. • Preprocessed 1350+ multispectral images to 256x256 dimensions and generated corresponding CDL masks, improving mask quality through Dilation, Opening and Closing techniques.• Trained a UNET-based segmentation model using RGB, Near Infrared band and NDVI index to accurately classify wheat and non-wheat pixels, achieving 94% classification accuracy.

Islāmābād, Pakistan

• Conducted detailed analysis on a dataset exceeding 100,000 data points to derive insights regarding CO2 emissions.• Identified and correlated pivotal factors influencing CO2 emissions, providing valuable information for targeted environmental strategies.• Developed and implemented advanced Machine Learning (ML) and Deep Learning (DL) models to automate the data collection process. This led to a significant enhancement in the accuracy of measuring R22 gas levels in Air Conditioning units, with Multivariable polynomial regression achieving a Mean Squared Error (MSE) of less than 1.8.

Islamabad, Pakistan

Teaching Assistant for Fundamentals of Computer Programming (CS-110)• Held 100+ office hours to teach 45 students fundamental Python concepts including Loops, Functions, Arrays.Teaching Assistant for Object Oriented Programming (CS-220)• Taught 45+ students introductory concepts relating to C++, including Classes. Inheritance and Polymorphism.

Islamabad, Pakistan

• Conducted comprehensive literature reviews to inform research direction.• Developed and implemented a General Regression Neural Network (GRNN) model to predict IoT immeasurable water quality parameters like 5 Day BOD, Nitrates and Phosphates.• Utilized measurable factors such as pH, Temperature, Total Dissolved Solids, and Electric Conductivity for accurate predictions.• Enabled real-time monitoring of Islamabad’s local reservoir's water quality, contributing to real-time environmental insights.

• Assisted 3 Computer Vision Specialists to develop an object tracking module for autonomous vehicles using YOLOV5, to achieve 95% test accuracy on the experimental video dataset.• Optimized company resources by overlooking GPU acquisitions for workstations, resulting in 30% cost reductions and maintaining 100% of workstations’ processing capabilities.

Islāmābād, Pakistan

Bachelor Of Science - Bs, Computer Science, 3.73/4.0

O/A Levels, Mathematics And Computer Science, 9 A*S, 2As