Zaraar Malik

Zaraar Malik LinkedIn : https://www.linkedin.com/in/zaraar-malik GitHub : https://github.com/Zaraar125 Email :-Mobile : - Education • FAST National University of Computer and Emerging Sciences Islamabad, Pakistan Bachelor of Science in Artificial Intelligence ; CGPA: 3.59 2021 - 2025 Experience • MWAN Mobile Rawalpindi, Pakistan ML Intern Feb 2025 – Apr 2025 ◦ Customer Service Chatbot: Improved quality of response given to customer by using Retrevial-Augmented-Generation(RAG) Chatbot by extracting company information embeddings stored in ChromaDB database which is retrevied later to improve LLMs answer given to the customer. • AIO – Silicon Valley Stealth Startup • Artificial Intelligence & Machine Learning Lab Islamabad, Pakistan AI Intern June 2024 – Aug 2024 ◦ BERT-Based Entity Extraction for Restaurants: Improved restaurant owners’ ability to detect genuine issues in customer reviews by preparing a custom entity extraction dataset, fine-tuning BERT on entity-specific segments, and designing a metric to evaluate extraction accuracy. Islamabad, Pakistan AI Intern June 2023 - Aug 2023 ◦ AI-Powered Video Dubbing System: Enabled multilingual dubbing with accurate lip-syncing by integrating Whisper for audio transcription, Coqui and Tortoise TTS for voice generation, and Wav2Lip for realistic speech-to-video alignment. ◦ Neural Audio Style Transfer: Replicated timbre and texture in audio by implementing CNN-based Neural Style Transfer techniques, enabling creative cross-domain applications in sound design. Projects • Snap Shop - Multi-Garment Mapping with Diffusion Models: Designed an easier way for people to gauge themselves with multiple fashion items by building a web-based platform that generates realistic images of users wearing selected garments. Empowered customers and designers to visualize fashion items on real people and create AI-generated model visuals and video ads. Achieved this by integrating multi-garment conditioning, and fine-tuning diffusion models using LoRA on custom datasets. • Procedural Game Level Generation: Trained DCGAN and WGAN models using the Mario Game Levels dataset to generate Super Mario-style game levels, enabling AI-based procedural content creation for game design. • Automating Hidden Object Games with Object Detection: Automated the process of detecting hidden objects in cluttered scenes using YOLO and Faster R-CNN. I fine-tuned the models on datasets like iMaterialist and Object365, then developed a frontend interface for users to upload images, trigger detection, and view results including detected objects, score, and inference time. • Multi-Modal Meme Classification: Built a classification model that analyzes meme images and accompanying text to predict sentiment (Positive, Negative, Neutral) by designing dual-input layers for vision and language, then merging them into a unified architecture after early fusion. Programming Skills • Languages & Frameworks & Deployment: Python, C++, JavaScript, PyTorch, Scikit-Learn, Keras, FastAPI, Flask, Docker, Kubernetes, Vercel, GCP. • Deep Learning: CNNs, LSTMs, BERT, RAG, Wav2Lip, GAN, ViTs, Transformers, YOLO, SAM, Diffusion. Honors and Awards • Data Quest, Nascon 2023: Winner of Data Quest, the biggest AI Hackathon of Pakistan held in Nascon. • AIO Hackathon, FAST 2023: 2nd Position achieved in AI-Hackathon conducted by AIO in FAST. • DEAN’s List: Included in Dean’s List of Honors (4 times) for excellent academic performance.