Muhammad.
Kanvis
Muhammad Hammad Yousaf

Muhammad Hammad Yousaf

Computer Science Student & AI Researcher specializing in Computer Vision and Edge Deployment
Hire Me

Overview

A Computer Science student at LUMS with extensive experience in developing AI-powered applications, including RAG systems and compressed LLMs for edge devices. He has a proven track record in computer vision research, specifically in image deblurring and wildlife monitoring systems.
Lahore, Pakistan

Experience

Jun 2025 – Aug 2025

INTERNSHIP

Summer intern

Computer Vision Lab (LUMS)
Lahore, Pakistan
Developed a U-Net–based rain-aware preprocessing pipeline integrated with MegaDetector, using tunable confidence thresholds to improve animal detection performance under rainy conditions achieving almost 100 % false positives avoidance. Fine-tuned MegaDetector on the COD10K dataset to improve detection by ~5% in camouflaged animal images. Optimized the pipeline for deployment on remote edge devices for wildlife monitoring and alert systems by compressing the model to 16.58 Mbs
•Developed a U-Net–based rain-aware preprocessing pipeline integrated with MegaDetector, using tunable confidence thresholds to improve animal detection performance under rainy conditions achieving almost 100 % false positives avoidance.
•Fine-tuned MegaDetector on the COD10K dataset to improve detection by ~5% in camouflaged animal images
•Optimized the pipeline for deployment on remote edge devices for wildlife monitoring and alert systems by compressing the model to 16.58 Mbs

Key Achievements

→Achieved almost 100% false positives avoidance in rainy conditions.
→Improved detection by ~5% in camouflaged animal images.
→Compressed model to 16.58 Mbs for edge deployment.
U-NetMegaDetectorCOD10K datasetEdge DevicesComputer Vision

Jun 2024 – Aug 2024

INTERNSHIP

Summer intern

Mindstorm Studios
Collaborated with a multidisciplinary team to develop a hyper-casual game ArroAce using Unity and C#. Submitted the game for the annual Summer Game Jam Competition where it received positive feedback
•Collaborated with a multidisciplinary team to develop a hyper-casual game ArroAce using Unity and C#.
•Submitted the game for the annual Summer Game Jam Competition where it received positive feedback

Key Achievements

→Received positive feedback at the annual Summer Game Jam Competition.
UnityC#

Project Portfolio

ACADEMIC

Present

AI-Powered Note Taking App

Full Stack Web Developer

Developed a full-stack note-taking application using React, Flask, MongoDB, and REST APIs, integrating AI-assisted features through LLM APIs. Deployed the application on Vercel, achieving sub-2-second query times and implemented automated UI testing using Selenium.
ReactFlaskMongoDBREST APIsLLM APIsVercelSelenium
ACADEMIC

Present

LUMS Academic Knowledge Assistant

Retrieval-Augmented Generation Developer

Designed and implemented an end-to-end RAG system with data ingestion and retrieval pipelines using LangChain, ChromaDB, and BGE embeddings to process university documents. Integrated LLaMA 3.3 70B model for response generation with multi-turn dialogue, achieving sub-4-second query latency, containerized with Docker and deployed with a Streamlit frontend The web application makes the process of understanding university documents much easier by simply talking to a chatbot.
LangChainChromaDBBGE embeddingsLLaMA 3.3 70BDockerStreamlit
ACADEMIC

Present

Kernel Based Image Deblurring

Senior Project Lead

Researched and implemented kernel-based image deraining techniques to improve object detection under adverse weather conditions. Modeled rain streaks as convolution kernels and tackled deraining as a convolution/deconvolution problem, using a multi-module pipeline and a mix of AI and computer vision strategies for increased efficiency
AIComputer VisionConvolution Kernels
ACADEMIC

Present

Llama 3.1_8b Compression

AI on Edge Devices Developer

Applied multiple pruning, quantization and outlier suppression techniques to the Llama 3.1 8B model. Utilized llama.cpp library and successfully compressed the model down to ~8% with minimal quality loss and deployed the inference-ready compressed model on Raspberry Pi 5
Llama 3.1 8Bllama.cppRaspberry Pi 5PruningQuantization
ACADEMIC

Present

Autonomous Road Navigation

Robotics Developer

Utilized the established PIDNet for road segmentation and used segmentation maps as input into a proportional controller that outputted control commands based on scene information. Achieved high inference rates while processing 30 FPS input streams. Simulating real world scenarios
PIDNetProportional ControllerAI for Robotics

Education

Sep 2022 – Jun 2026

Bachelor of Science

Lahore University of Management SciencesSpecialization in Computer Science

Relevant Coursework

Software EngineeringDatabase SystemsData ScienceMachine LearningDeep LearningAI on edge devicesAI for RoboticsComputer Vision

Impact & Recognition

Professional Certifications

Amazon Web Services (AWS) · 2026

AWS Cloud Practitioner Essentials

Social Impact & Activities

Present

Wildlife Instagram Page

Manager

Manage a wildlife-focused Instagram page documenting and discussing local wildlife and conservation.

EngagementORGANIZER
Ajoka TheatreOTHER

Stage Productions

Performed in stage productions with Ajoka Theatre and acted in independent student films.

Skills & Interests

Python

LangChain

RAG

MERN

Git

PyTorch

C++

llama.cpp

Unity

C

C#

SQL

Selenium

Docker

MongoDB

RoboFlow

TensorFlow

.NET

TVM

AWS

ROS

Figma

Let's Connect

I'd love to hear from you. Feel free to reach out.

These unlock once you finish onboarding and publish your profile.

More people on Kanvis

Browse more people