Muhammad Tayyab

AI Engineer specializing in end-to-end Machine Learning and Agentic RAG systems.

Based in: Karachi, PakistanMost recently: Web Development Intern, MUET FM 92.6

Award-winning AI Engineer with a 3.6 CGPA and expertise in architecting scalable RAG systems, LLM fine-tuning, and real-time computer vision solutions. Proven track record of building production-ready applications, including an AI-powered fitness startup and high-accuracy exoplanet classification systems.

Experience

Web Development Intern

MUET FM 92.6internship

Spearheaded full-stack development of the official website, migrating legacy systems to a modern React architecture. Optimized frontend performance, reducing page load times by 40% and boosting SEO rankings through server-side rendering (SSR) best practices.

ReactSSR

Skills

Technical

LangChain

XGBoost

RAG

LangGraph

OpenCV

Scikit-learn

LightGBM

TensorFlow

CNN

YOLO

AWS Bedrock

FastAPI

Flask

Streamlit

Pinecone

Docker

U-Net

FAISS

MongoDB Atlas

Time Series Forecasting

PyTorch

RNN

GitHub Actions

Digital Ocean

n8n

Model Fine-tuning

Next.js

CatBoost

Languages

Python

TypeScript

Java

Projects

Google Kaggle Competition

personal

Fine-tuned Gemma 2 using a Pakistan-specific image dataset.

Gemma 2

Fitro - AI-Powered Fitness App

Founder/Developerprofessional

Problem: In Pakistan, millions struggle with calorie tracking due to limited health literacy and app fatigue. Solution: Built a mobile app using computer vision to detect food calories from images and use LLMs for real-time fitness guidance. Impact: Enabled >90% faster logging of meals vs. manual entry. API-first LLM setup ensures affordability and scalability across devices.

Computer VisionLLMsAPI-firstGemma 2n8n

ExoQuark | AI-Powered Exoplanet Classification System

Lead Engineerpersonal

Problem: Astronomical datasets from Kepler, K2, and TESS showed >30% false positives in exoplanet classification. Solution: Engineered ExoQuark, an AI ensemble (XGBoost + LightGBM) achieving ∼89% classification accuracy across satellites with calibrated thresholds and explainable outputs. Impact: Cut false detections by 40%, boosting reliability of exoplanet research through an open-access platform for scientists.

Next.jsTypeScriptThree.jsAWS BedrockFlaskXGBoostLightGBMCatBoost

Visit the site View the code

Nexora | AI-Powered Voice Platform

Developerpersonal

Problem: Creating realistic, natural-sounding voice synthesis and custom voice cloning was complex and inaccessible for most developers. Solution: Built an AI-powered voice platform that transforms text into natural speech and enables custom voice clone creation with studio-quality realism using Chatterbox TTS, deployed serverless via Modal.

Next.jsshadcn/uiChatterbox TTSModal

Visit the site View the code

Zentrix | Agentic RAG Chatbot

Full-stack Developerpersonal

Problem: Users lack a unified system to query personal documents, notes, and tasks using natural language with reliable, source-grounded answers. Solution: Built a full-stack agentic RAG system where users upload PDFs and documents; an AI pipeline embeds content into a vector store, retrieves relevant chunks on query, and generates cited answers via Gemini Flash with real-time streaming. Impact: Enables source-cited Q&A over private documents with a built-in RAG vs. direct Gemini comparison mode, integrated todo management, and a mobile-ready API layer.

Next.js 15ConvexGoogle Gemini APITypeScriptTailwind CSSClerkVector StoreRAGDockerGitHub ActionsDigital OceanNode.js

Visit the site View the code

Education

Bachelor of Science in Artificial Intelligence (BSAI)

Mehran University of Engineering and TechnologyArtificial IntelligenceGPA 3.6Jul 2023 – Aug 2027

Diploma in Information Technology (DIT)

Computer Man Institute of Emerging TechnologiesInformation TechnologyOct 2021 – Aug 2022

Recognition

Awards

Winner, Technofest 2026 - ML Mastery Competition

Technofest2026

Built and optimized an ML model to predict life expectancy, achieving an R2 score of 0.97 through strong data preprocessing, feature engineering, and model evaluation.

Runner-Ups / Finalist: PROCOM 26 AI Grand Prix

FAST NUCES2026

0.85 recall, 0.8+ accuracy; 3 engineered features, no internet.

Finalist: IBA ProBattle 26, ML Module

IBA2026

0.41 mIoU on Duality AI synthetic data using U-Net + ResNet ensemble.

Finalist: Hacktober Hackathon 2025

MehranUET2025

0.50 recall on a heavily class-imbalanced dataset.

Finalist: PROCOM 25

FAST NUCES2025

0.99 F1 using SHAP top-25 features + custom ensemble.

Winner, NASA Space Apps Challenge Hyderabad (2025)

NASA Space Apps Challenge2025

Recognized for building "ExoQuark," an AI/ML space exploration tool. Nominated for Global Round.

Winner, AWS MUET Hacktoberfest (2024)

AWS / MUET2024

Developed a cross-platform mobile application using Expo React Native.

Winner, Progressive Programming 2024, Tech Arena

Tech Arena2024

Placed in offline core programming challenge testing algorithmic problem-solving under time constraints.

NASA Space Apps Challenge Winner

NASA

Won the competition with a 10/10 score from the judges for the ExoQuark ML model.

Hire MuhammadReach out about a role, a contract or a conversation.For recruiters

Muhammad's twin is AI, it can make mistakes.

More people on Kanvis

Browse more people Create your own Kanvis