Amina Ashfaq
Data Scientist | AI Engineer--
linkedin.com/in/amina-ashfaq06
Islamabad, Pakistan
Skills
Languages and Frameworks: Python, SQL, C++, R
ML Tools and Libraries: Pytorch, Tensorflow, Scikit-learn, Langchain, OpenCV, Diffusers, ComfyUI, LLMs/Agents
Data Skills: SQL Server, MongoDB, Chroma, Weaviate, Tableau, PowerBI, Pandas, Numpy, D3.js
Professional Experience
Instructor - National University of Computer and Emerging Sciences
02/2024 – present
● Developed interactive curricula for undergraduate courses in AI, Database Systems, and ML Labs, integrating
real-world applications and cutting-edge tools (e.g., Python, TensorFlow, SQL) to enhance student engagement.
● Designed hands-on coding exercises and case studies tailored to advanced topics, enabling students to master
complex concepts such as neural networks, database optimization, and predictive modeling.
● Facilitated workshops and lab sessions, introducing students to tools like Jupyter Notebooks, MySQL, and PyTorch,
which empowered them to pursue internships in tech driven industries.
Mentor - Genesys Lab
06/2025 – present
● Supervised and mentored a team of students during a 6 week summer internship, guiding them in applied research on
advanced technologies like LLMs,RAG and Computer Vision, using tools like Python, Hugging Face, and OpenCV.
● Led the development of an intelligent information system that automates opportunity alignment through AI-powered
semantic search, document parsing, and contextual recommendations, supported by dynamic simulations.
Projects
Image Personalization for Virtual Try-on Applications - Masters Thesis
● Built a Virtual Try-On solution to improve customer confidence and reduce product return rates.
● Applied Dual-LoRA fine-tuning on FLUX.1 Dev, preserving user identity and clothing variations, achieving strong
evaluation scores (DINO 0.5, CLIP 0.82).
● Optimized training pipeline on RTX 3090, cutting fine-tuning time by ~40%.
● Delivered composite image generation with multiple customizations while maintaining identity integrity.
Image Captioning with a Hybrid CNN-Transformer Architecture
●
●
●
●
●
Built an end-to-end image captioning system using a hybrid CNN–Transformer architecture for vision and NLP.
Combined ResNet50 for feature extraction with GPT for autoregressive caption generation.
Developed a custom PyTorch dataset pipeline to preprocess ~120K MS-COCO images with caption tokenization.
Trained the model with cross-entropy loss + teacher forcing, achieving stable convergence on GPU.
Evaluated captions using BLEU, METEOR, and CIDEr metrics for quantitative performance assessment.
Student Academic Performance Analysis & Predictive Modeling
● Conducted end-to-end data analysis on student attendance, assessments, and exam results using Python.
● Built predictive regression models (scikit-learn) to forecast exam scores and detect at-risk students early.
● Extracted insights on attendance–performance correlations using Pandas, Scikit-learn, Seaborn.
Education
National University of Computer and Emerging Sciences (FAST), Islamabad
Masters in Data Science - CGPA: 3.91
University of Sargodha, Sargodha
Bachelors in Information Technology - CGPA: 3.49
08/2023 – 06/2025
09/2017 – 06/2021
