Zara Zia

Academic

Verbatim text

Zara Zia
03345368027
House Number 27, Street 22, Sector C, Phase 1, DHA, Islamabad
LinkedIn:
Education
FAST NUCES, Islamabad
BS(AI)
Artificial Neural Networks, Cloud Computing, Computer Vision, Digital Image Processing, Knowledge
Representation and Reasoning, Machine Learning, MLOps, NLP
Benchmark School System
Physics, Maths, Urdu
Benchmark School System
Physics, Chemistry, Maths, Biology
Projects
Final Project:
Patronus (Nexmon, PyTorch, FastAPI, PostgreSQL, GIT, ML Flow, Apache Airflow, Jenkins, Docker)
A 5G-enabled system for real-time patient monitoring, which includes fall detection and vital signs monitoring without
any sensors or wearable devices, using just 5G.
Semester Projects:
Aqua Debris (Resnet, OpenCV, Pillow)
Detecting Marine Debris using a customized Attention Activated Residual U-Net.
SemiCol challenge (TensorFlow, OpenCV, Scikit)
Semi-supervised learning for colorectal cancer detection from whole slide images.
Speech to Speech ChatBot (TensorFlow, NLTK, TFX)
Voice-Enabled Closed-Domain Chatbot with Knowledge Graphs and NLP.
Work Experience
AI/ML Intern, One Network
May 2023 - July 2023
Automated Vehicle Classifier for the Motorway.
FullStack Developer , PCN Lab (FAST NUCES)
August 2023 - present
Working on development and deployment of Patronus.
Research Partner, JAZZ 5G Lab (NUST SINES)
October 2023 - present
Collecting and analyzing data for Patronus, for contributing findings to scientific publications.
Skills & Tools
Professional Skills:
Agile Adaptability, Project Management, Problem Solver, Collaborative Leader
Technical
Skills:
Wireshark, Nexmon, Kubernetes, Docker, GIT, Jenkins, AWS Cloud, ML Flow, Apache Airflow, Ngrok,
Flask, HTML, CSS, React, SPARQL, MySQL, Node JS, Express JS, ML/DL Algorithms
Achievements
3x Winner of Parwaz-e-Takhayyul, Tech Ideathon, by Google Developer Student Club (2022, 2023, 2024)
Training / Certification
AWS Cloud Practitioner
Activities
Creative Lead, Artificial Intelligence and Data Science Society (2022 – 2023)
Events Lead, Fast Society of Arts (2022 - 2023)
Head of Artificial Intelligence, IEEE FAST (2022 - 2023)
Interests
Debating, Reading, Sports
zaraziaofficial@gmail.com
https://www.linkedin.com/in/zara-zia-a5b761212/
Majors:
Cambridge International A-levels (
)
Cambridge International O-levels (
)
The AI Modeling Assistant (AIMA)
In the fields of gaming, animation, publishing, and art, artists face challenges with concept
modeling and turnaround sheets, often leading to wasted effort and resources on designs
that don't progress. The AI Modeling Assistant (AIMA) offers a solution by using
generative models to minimize sketching, producing varied poses from few initial designs,
enhancing productivity, and enabling customization. AIMA streamlines the design process,
addressing key issues in early visual storytelling.
Main Features:
1. Object Integration:
● Interactive interface for user-driven masking to select the area of interest
● Option to enter a text prompt for custom object integration.
● Algorithm-driven integration to ensure natural placement of objects.
2. Image Refinement:
● Automatic refinement of input images.
● Optional text prompt for specific refinement instructions.
3. Pose Generation:
● Generate distinct poses (front, back, right-side, left-side) based on text
prompt.
● Consolidate all poses into a single character sheet for easy reference.
Technology Used:
Python, FastApi, HTML/CSS/JS, Hugging-
face Transformers, Github
Supervisor Name:
Mr. Shoaib Saleem
Group Members:
Mohammad Osama (i20-0931)
Ana Riaz (i20-2483)
Adeen Hassan (i20-0916)
AutoCruiter
AutoCruiter aims to revolutionize the recruitment process at an enterprise level, leveraging the
power of AI to make hiring more efficient, unbiased, and effective. At its core, it simplifies the
recruitment workflow by automating several key phases, from CV shortlisting to conducting
preliminary interviews. Its innovative technology suite is designed to identify the most suitable
candidates for a position based on their resumes and to generate personalized interview
questions that delve into each candidate's unique qualifications and experiences. Additionally,
the system employs audio-visual emotion recognition during interviews to gain deeper insights
into candidates' responses. This blend of automation and AI-driven analysis offers a scalable
and customizable solution that not only speeds up the hiring process but also ensures a higher
quality of candidate selection, all within a user-friendly interface.
remote automated interviews, speeding up the hiring process and reducing biases, but
highlights the need for multilingual support to fully engage the local talent pool.
Harnesses LangChain for NLP, cosine similarity for CV matching, and advanced AI for
personalized interviews and emotion analysis, all within a scalable, cloud-based architecture
for a seamless hiring experience.
Technology Used:
Langchain, Django, OpenCV, MySQL
Workbench, FAST API
Supervisor Name:
Dr. Akhtar Jamil
Group Members:
Zainab Shah (i20- 0478) 03316122294
Rubab Fatima (i20 - 0774) 03305217185
Ayesha Zaheer (i20 - 0974)03335601459
E-Learn
An innovative eLearning platform that leverages AI technology to enhance the learning
experience. Separate dashboard for students and teachers. Teachers have the ability to
create courses and upload relevant videos. Students, upon enrollment, are presented with
their respective courses on their dashboard. Features:
AI generated Initial knowledge assessment for students.
Based on the assessment answers, AI generated helping material for students.
Automated Powerpoint Slides generation to present AI generated helping material.
Throughout the course, students can take comprehensive assessments to gauge their
understanding.
Text-to-speech that mimics the teacher's voice.
Technology Used:
Python, Pytorch, Chroma DB, HTML, CSS,
JavaScript, MongoDB
Supervisor Name:
Mr. Umair Arshad
Group Members:
Saleh Ahmad (i20 - 0605) 03120579200
Usman Naveed (i20 - 0826) 03328025177
Muhammad Ammar (i20 - 0783)
03174331619
TECH TUTOR
Revolutionizing Computer Science education by addressing traditional limitations. Limited
guidance, rigid schedules, and minimal interaction. Our mission is to enhance learning
through personalization and accessibility with the help of AI Tutor, shaping the future of
education. Enhance student learning experiences through personalized, interactive, and
adaptive techniques.
Developed AI-powered tutor for personalized e-learning with real-time voice and facial
expression. Utilized fine-tuned LLM (Llama 2-7b) for Q&A interaction. For lip-syncing and
getting real time face to communication settalker is being used.
Major Features include:
● Voice to voice chatbot
● Face to face communication
● Quiz Integration and progress tracking
● Personalized question answering
Technology Used:
Python, LangChain, OpenCV,
Node JS, Flask
Supervisor Name:
Dr. Mehreen Alam
Group Members:
Ans Hussain (i20-1824) - 03093301981
M. Shaffay (i20 - 2391) - 03152819002
Syed Zulfiqar (i20 - 2337) - 0333511665
Familiar
Familiar is the customized sentient Avatar for grief support. After losing a loved one, most
people go into a destructive spiral of depression, stress and grief which highly damage their
mental health but they don't get into therapy with the mindset that the pain will go away and
they can handle it themselves. We tend to push such people toward treatment so they
should select the therapy instead of living their lives in a closed tunnel of disbelief and
trauma. We have implemented the deep learning techniques and used Deep Fake for more
realistic Avatar. User will sign up first and upon signing up will provide the picture and voice
snippet. Our model will create the avatar with response from LLM and voice will be cloned
from the voice provided by the user. For more realistic approach we have used lip sync and
expression sync models for Avatar.
Models Used:
-
STT - silero
-
TTS - openvoiceclonning
-
LLM - mixtral 8x7b
-
DEEP FAKE – sadtalk
Features Include:
-
Chatbot: Giving answers to the questions of the user using LLM
-
Chat history of the user
-
Avatar
-
Relationship with the Avatar for the user to chat within the context
Technology Used:
Python, PyTorch, React, FastAPI
LangChan, Visual Studio
Supervisor Name:
Mr. Shoaib Saleem Khattak
Group Members:
Faian Mehmood (i20 - 2704)
+923026602996
Syed Addan Hafeez (i20 - 0818)
+923314844004
Hassam Nazir (i20 - 2438)
+923185321325
GraphDiag
An e-health dashboard aimed at assisting doctors in diagnosing patients' diseases,
forecasting ICD-9 codes, and assessing mortality rates using Explainable AI. Predicting the
ICD-9 Codes using Clinical Notes, Lab Reports from MIMIC-III dataset and providing
explainability using glass-box testing and Large Language Models. A medical chatbot is
integrated that provides the following features:
• Query over the Patient Records
• Consult the doctor regarding any medical query
A predictive analysis using the patient sequential diagnosis information is used for predicting
the ICD-9 code. This uses the knowledge graphs methodology for maintaining the
sequential records. Predict patient mortality utilizing comprehensive clinical data for
enhanced healthcare decision-making.
Technology Used:
Python, Flask, Scikit-learn, SQLite, Mistral-
AI, Knowledge Graphs,  Jupiter Notebook
Supervisor Name:
Dr Amna Basharat
Group Members:
Abdullah Hameed (i20 - 1799)
0308-0000799
Hassan Kamran (i20 - 2435)
0304-5166261
InspectAI
InspectAI is a groundbreaking project at the forefront of inspection technology
innovation. By leveraging smart spectacle technology and advanced computer vision
algorithms, InspectAI aims to revolutionize inspection processes across various
industries. The project focuses on empowering inspectors with real-time insights,
precise anomaly detection, and accurate dimension measurements, ultimately
enhancing efficiency, accuracy, and user experience. With a modular system
architecture, cutting-edge object detection models, and intuitive user interfaces,
InspectAI sets out to redefine the standards of inspection and quality assurance.
Key features include user authentication, real-time video processing, data
management, security measures, user-friendly interfaces, and system integration.
InspectAI sets out to enhance efficiency, accuracy, and user experience in inspections
through innovative technology and comprehensive functionalities.
Technology Used:
Python, Docker, Google Spectacles,
OpenCV, Deep Learning, Github, AR
technologies
Supervisor Name:
Mr. Hassan Raza
Group Members:
Abbas Mustafa (i20 - 2404)
Cell #: 03117866600
Ali Chohan (i20 - 0884)
Cell #: 03315658243
Shayan Bakht (i20 -1753)
Cell #: 03109070343
JungleSurveil
JungleSurveil is an innovative Deep Reinforcement Learning (DRL) algorithm designed for
efficient path planning in jungle environments. The algorithm is specifically tailored for
Unmanned Aerial Vehicles (UAVs) tasked with surveillance missions in complex terrains.
The primary objective of JungleSurveil is to optimize UAV path planning to prioritize high-
priority areas within the jungle while ensuring comprehensive coverage of the entire terrain.
JungleSurveil utilizes a combination of DRL techniques and image processing algorithms.
It takes input in the form of aerial images of the jungle, with high-priority areas marked.
Through iterative learning and exploration, the algorithm dynamically adjusts the UAV's path
to maximize coverage of both high-priority and non-priority areas while minimizing
redundant scanning. Extensive simulations have demonstrated the effectiveness of
JungleSurveil in optimizing path planning for jungle surveillance missions.
Technology Used:
TensorFlow, OpenCV, Python,
Tkinter, VS Code
Supervisor Name:
Dr. Ahmad Din
Group Members:
Ahmad Masood (i20 - 1754)
0340 7904113
Faizan Aslam (i20 - 1843)
0320 7080450
M. Qasim Khan (i20 -1878)
0349 6496909
MediaSynthX
This project aims to develop a web-based user-centric application that utilizes generative
AI techniques and state-of-the-art models for the generation of multimedia content,
including videos with audio. The system's core objective is to simplify the creation of
visually compelling videos from textual descriptions.
The project's key technical objectives involve the integration of generative AI models for
text-to-video and text-to-audio transformations then combining both into one video.
Additionally, the project incorporates emotional elements into the audio, ensuring that the
sound aligns with the desired emotions conveyed in the video. This enhancement adds a
nuanced layer to the multimedia content, making it more engaging and resonant with
diverse audiences. By automating the video generation process, the project enhances
efficiency and encourages creativity. The system targets diverse users, including content
creators, educators, marketers, and travelers. It promotes innovation and scalability,
empowering individuals and professionals to easily create engaging videos for various
purposes.
The key feature is that user can input a textual description of the video they want to
generate, and the video generated from it will be rendered on the screen.
Technology Used:
Python, Generative AI, Flask, Stable
Diffusion Models, Coqui, Colab, MongoDB
Supervisor Name:
Mr. Hassan Raza
Group Members:
Eysha Raazia (i20 - 1818)
Cell No: 03409384560
Alishba Laeeq (i20 - 2459)
Cell No: 03325437609
Ayesha Qureshi (i19 - 1700)
Cell No: 03017703144
PoseQuest: Efficient Pose Retrieval from Large Motion
Databases
A research and development initiative aimed at improving pose retrieval methods from
large motion databases. We explore a range of established techniques, including cosine
similarity, k-nearest neighbors (KNN), KD-Trees, and Dynamic Time Warping (DTW),
assessing their performance based on metrics like Principal Component Analysis (PCK),
Mean Per Joint Squared Error (MPJSE), Mean Squared Error (MSE), and computational
time (seconds). Application of a sliding window technique to the nearest neighbor search,
designed to potentially reduce retrieval times. The windowing approach is not on the
features, as traditionally done, but directly on the pose neighborhoods  themselves. This
innovative strategy has shown promising results in our preliminary trials, potentially
streamlining the retrieval process and ensuring more rapid access to relevant motion data
without sacrificing accuracy.
Features include:
- Efficient 3D Motion Retrieval, 3D Motion Visualization, and evaluation on metrics such as
time (seconds), Mean Per Square Error (MSE), Mean per Joint Square Error (MPJSE),
and Percentage of Correct Keypoints (PCK).
Technology Used:
Python, Matlab, Flask, HTML, CSS
Visual Studio Code, Jupyter Notebook
Supervisor Name:
Mr. Shoaib Saleem Khattak
Group Members:
Areeba Ayaz (i20 - 0460)
Muhammad Bilal (i20 - 1877)
Asna Muzafar (i20 - 1805)
Patronus - Remote Healthcare Using WiFi Sensing
The Patronus project is a pioneering venture in healthcare and safety monitoring,
distinguished by its non-intrusive approach and reliance solely on Wi-Fi signals for data
acquisition. Unlike traditional methods, Patronus doesn't require any wearables or additional
devices, ensuring a seamless and hassle-free experience for users. By harnessing Channel
State Information (CSI) data from Wi-Fi signals, Patronus can detect falls, analyze heart
rates, assess breathing rates, and determine an individual's state with exceptional accuracy.
What sets Patronus apart is its ability to analyze motion states using only the disturbances
in signals, distinguishing between walking, sitting, standing, and lying positions. At its core,
Patronus employs an ensemble model structure, merging multiple algorithms to achieve
precise and real-time detection of abnormal vital activities, all without intruding on the user's
daily life. This innovative approach heralds a significant advancement in digital health
monitoring and safety, envisioning a future where healthcare is effortlessly accessible and
seamlessly integrated into everyday living. Patronus represents a commitment to enhancing
healthcare and safety through innovative, non-intrusive, and Wi-Fi-based monitoring
solutions.
Technology Used:
Python,FastAPI, Wireshark, Nexmon
CSI, Tensorflow,jenkins, Pytorch, React,
JavaScript,PostgreSQL,MLflow
Supervisor Name:
Dr. Arshad Islam & Dr. Usman Habib
Group Members:
Masham Zahid  (i200572)
Faryal Fatima Sabih  (i200685)
Zara Zia (i200471)
CogniVision: Redefining Video Interaction for Masses
Introducing a groundbreaking solution: an inclusive educational video interaction system
designed to revolutionize access to knowledge for individuals with limited mobility or visual
impairments. Say goodbye to traditional barriers and hello to seamless engagement with
educational content through natural language queries, sound integration, and chat with
video.
Our project tackles the critical accessibility gap in educational video content, prioritizing
inclusivity, voice-controlled interaction, and accessibility features. By empowering users
regardless of physical abilities, we bridge the divide, offering an equitable and enriching
educational experience.
But the impact doesn't stop there. Our system's applications extend far beyond education.
Imagine unlocking a world of entertainment where everyone can fully participate, or
enabling data scientists to effortlessly analyze and interact with video content in real-time.
With features like runtime segmentation, title generation, and voice/textual interaction, the
possibilities are endless.
Technology Used:
Python, OpenCV, Vue.js, Git,
HuggingFace, Pytorch, Django, Apache
Airflow.
Supervisor Name:
Dr. Hassan Raza
Group Members:
Ubair Nisar (20i-2348)
+92 309 6128600
Amina Bibi (20i-1867)
+92 302 1116120
Muhamad Ali Shah (20i-1836)
+92 305 1493055
In-VERSE: Gateway to virtual realms from reality
In-VERSE is a Computer Vision driven 360-degree environment generation solution that
empowers real estate developers, and hotel owners to provide a more complete and
immersive viewing solution for their customers. We let the users to take pictures of a
particular room that they want to view virtually, and our model generates a 360x180-degree
view of the room allowing for remote exploration.
From the application point-of-view; the app has two major ends:
1. The user / visitor end: Where the visitor can view pictures of the room, and can engage
in two other more immersive viewing properties:
⚫
View the room as a 3D Layout.
⚫
View the room in a Virtual Reality Headset.
2. The admin / host end: Where an agent uploads the pictures of a property.
Core Features that promise feasibility:
1. No special cameras required; the model can work without camera’s metadata involved.
2. Completes the entire image -> 3D Layout process within ten minutes on M1 Chip / Low-
Cost Consumer GPU
3. Highly accurate estimations on Manhattan styled room layouts.
4. Low-cost VR Headsets required for VR-based viewing.
Technology Used:
Python, PyTorch, OpenCV, SciPy, Hugging
Face,
Open3D
Supervisor Name:
Mr. Umair Arshad
Group Members:
Maaz Ali Nadeem (i20 - 0452)
Shehryar Sohail (i20 - 0429)
Abdulwadood Waseem (i20 - 0988)
Frame Shift
FrameShift is a groundbreaking project in the world of 2D animation, utilizing the innovative
DragGAN tool to revolutionize the animation process. Unlike traditional methods that are
labor-intensive and time-consuming, FrameShift aims to streamline animation creation by
leveraging the power of Generative Adversarial Networks (GANs). By combining the
precision of traditional animation tools with the dynamic capabilities of DragGAN,
FrameShift offers animators unparalleled creative freedom and efficiency. The project is
divided into two key modules, each offering unique functionalities to enhance the animation
workflow. Module 1 focuses on augmenting and editing images using DragGAN, enabling
animators to generate frames with various manipulation types quickly. This module aims to
speed up the animation completion process. Module 2 is designed to animate characters
within images, incorporating movements like smiling, blinking, and face rotation. Both
modules are user-friendly and accessible, making the animation process more efficient for
animators of all skill levels. FrameShift's core features include the ability to manipulate
angles, features, and elements of a 2D image, providing animators with unprecedented
control over their creations. The project also aims to address the challenges of traditional
animation, such as time constraints and repetitive tasks, by offering a more efficient and
dynamic workflow. By bridging the gap between human creativity and artificial intelligence,
FrameShift is set to revolutionize the animation industry, empowering animators to bring
their ideas to life in ways never possible before.
Technology Used:
Python, PyTorch, Gradio, Nvidia, Visual St
udio
Code, GitHub, Docker
Supervisor Name:
Mr. Umair Arshad
Group Members:
Aamish Rafique (i20-0511)
Fateh Aayan (i20-0627)
Arooba Ali (i20-0776)
Atrial Fibrillation Detection Through Fingertips
Atrial fibrillation is a heart disease which can lead to major problems like strokes and heart
attacks. An app is being made that can detect AFIB patients. The user’s pulse is measured
using a pulse sensor that is connected to Arduino. This Arduino send the data to the host
where we run ML models on it. ML models are used to predict the ECG signals and pulse rate
through the pulse signals. These three features are then used to detect AFIB. The Application
is integrated to the models and Arduino board using Ngrok. In the Application, after the user
logs in, he is asked to place his finger on the sensor for two minutes. After two minutes the
AFIB detection result is displayed.
Features:
-AFIB result is displayed.Graphs of PPG, ECG and Respiratory Rate are displayed
Technology Used:
C++, Arduino, Pulse Sensor, Python,
Tensors, OpenCV,Flask, SQL Lite
Supervisor Name:
Dr.Uzair Iqbal
Group Members:
Danial Sadiq (i19 – 1871)
Ijaz Ahmed(i19 -1873)
Salman Saeed(i19 – 1763)
SpaceSculptor AI
SpaceSculptor AI represents a cutting-edge interior design application tailored to the needs
and desires of our users. Leveraging the transformative capabilities of artificial intelligence,
our platform redefines interior spaces into personalized sanctuaries of both aesthetic allure
and practical utility. With SpaceSculptor AI, users are empowered to conceptualize their ideal
living environments and witness them materialize before their eyes.
At the heart of our application lies a seamless process: users provide inputs including images
of their space, style preferences, and any specific requests. Armed with this information, our
advanced AI model sets to work, skillfully redesigning the interior to reflect the user's vision.
Moreover, SpaceSculptor AI fosters an interactive experience, enabling users to engage in
dialogue with the application for real-time feedback and refinement. Through this dynamic
exchange, our AI model adeptly replaces and removes objects as per the user's directives,
ensuring a truly customized outcome.
In essence, SpaceSculptor AI transcends the conventional boundaries of interior design by
placing the user at the forefront of the creative process. By seamlessly blending AI capabilities
withuser-centric features, our application empowers individuals to transform their living spaces
into personalized havens that seamlessly merge beauty with functionality.
Features include:
•
2D Image Generation
•
Conversational Component
•
Removal and Replacement of Objects
•
Design Rendering
Technology Used:
JPytorch, Tensorflow, Github, MongoDB,
Express, NodeJS, ExpressJS,
Python,Visual
Studio, NGROK, Google Colab
Supervisor Name:
Dr. Naveed Ahmad
Group Members:
Eisha Farrukh (i20 - 0657)
Khadija Irfan (i20 - 0803)
Khaula Atiq (i20 - 0765)

AI enrichment