Muaviz Maroof Khan

Forward-thinking Software Engineering student at DTU with a unique blend of core technical proficiency and enterprise Go-To-Market (GTM) experience. Proven track record in building advanced LLM/RAG architectures and deploying automated Agentic AI workflows that directly enhance sales productivity and outbound campaign engagement. Combining deep skills in data science, predictive modeling, and market intelligence, I excel at bridging the gap between complex engineering workflows and strategic corporate growth.

• Amazon ML Challenge 2024: Secured All India Rank (AIR) 377 in a high-stakes national machine learning competition, competing against thousands of top engineering talents across India.
• Competitive Programming: Solved 250+ algorithmic coding problems across premier platforms like LeetCode and CodeChef, demonstrating strong data structures and algorithms (DSA) proficiency.
• Leadership & Class Governance: Appointed as Class Representative (1st Year), driving cross-functional student coordination, peer conflict resolution, and seamless department-wide communication.

1.) NPTEL - Indian Institute of Technology Madras - The Joy of Computing Using Python - July-October 2024.

2.) NPTEL - Indian Institute of Technology Madras - Social Networks - January-April 2026.

3.) CodeWithHarry - The Ultimate Job Ready Data Science Course - July-January 2026

Intelligent RAG Learning Assistant 

Python, Llama 3.2, BGE-M3, OpenAI Whisper, Flask, FFmpeg, Pandas, NumPy | April 2026

• Architected an End-to-End Multimodal RAG Pipeline: Developed a 5-stage ingestion framework that processes unstructured video lectures into a time-indexed knowledge base using FFmpeg and OpenAI Whisper ASR for temporal transcription.
• Optimized Semantic Retrieval Layer: Implemented a high-precision bi-encoder layer using BGE-M3 and vectorized Cosine Similarity computations, eliminating keyword-miss failures to deliver sub-second, intent-aware concept retrieval.
• Contextual Synthesis Workflows: Deployed a Flask web application featuring prompt-engineered Llama 3.2 inference pipelines to synthesize fragmented transcripts into structured, pedagogically aligned explanations mapped to exact video timestamps.

Automobile Predictive Failure Detection 

Python, Scikit-learn, SMOTE, RandomizedSearchCV, Joblib, Pandas, Seaborn | June 2025

• Engineered Leakage-Free Data Pipelines: Built a robust predictive maintenance pipeline using industrial multi-sensor telemetry, strictly isolating SMOTE oversampling and data scaling to the training split to preserve real-world failure distributions.
• Evaluated & Serialized Champion ML Models: Benchmarked 7 diverse machine learning architectures (tree-based, kernel-based, and neural networks) using RandomizedSearchCV to optimize F1-Score and ROC-AUC.
• Productionized Deployment: Serialized the final inference stack into production-ready Joblib artifacts, enabling proactive manufacturing downtime prevention and telemetry tracking.

Traffic Sign Classifier (CNN Model) 

Python, TensorFlow, Keras, OpenCV, Scikit-learn, PIL, Matplotlib | June 2024

• Developed Deep Learning Architectures: Built and trained a Convolutional Neural Network (CNN) using TensorFlow/Keras, achieving 95% classification accuracy across 43 categories using the GTSRB dataset (50,000+ images).
• Optimized Computer Vision Preprocessing: Engineered an end-to-end preprocessing pipeline with OpenCV for image normalization, resizing, and colour-space transformations to maintain model stability under motion blur and low lighting.
• Implemented Diagnostic Frameworks: Formulated a Scikit-learn confusion matrix framework to isolate and analyze misclassifications, validating model reliability for Advanced Driver Assistance Systems (ADAS).