Hello World, my name is

Muneeb Ur Rehman Siddiqui.|

I am an Artificial Intelligence Engineer with hands-on experience building and deploying end-to-end AI applications across Computer Vision, Natural Language Processing, and Big Data Analysis.

Leveraging a strong foundation in designing data pipelines and fine-tuning machine learning frameworks, I also focus on developing production-oriented systems using tools like FastAPI and Docker to integrate intelligent solutions directly into user-facing applications.

Fully committed to building systems that bridge the gap between complex models and real-world utility.

Capstone Projects

Download my latest CV

01. About me

Hello, I'm an Artificial Intelligence Engineer based in Karachi, Pakistan.

I've always been fascinated by complex systems, whether they exist in technology, nature, or the fictional worlds. The stories that resonate with me most are those that leave no stone unturned, where every detail, connection, and consequence matters.

That same curiosity is what drew me toward Artificial Intelligence. I enjoy exploring how data, algorithms, and software come together to create systems that are greater than the sum of their parts. Rather than simply using existing tools, I'm most interested in understanding how they work and how they can be pushed further.

I am particularly drawn to projects that involve experimentation, research, and discovering new approaches to difficult problems. Whether it's Computer Vision, Data Science, or emerging developments in AI, I enjoy learning how ideas evolve and where the field is heading next.

I remain optimistic about the future of AI and the opportunities it creates, believing that today's research can become tomorrow's everyday reality.

Hard Skills:

Currently Exploring:

02. Where I have worked

BUKC ACCP
Arts Team Head
Private Tutor
Grades 5-11

Arts Team Head

@ BUKC ACCP Club

2026 - Present

Leading the creative initiatives and visual identity for the Bahria University Karachi Campus (BUKC) ACCP Club, driving student engagement through impactful design and event coordination.

Creative Direction & Event Management:
Oversaw all creative direction and ensured the successful, timely delivery of university events and club activities.
Team Collaboration:
Coordinated with club members to conceptualize and execute visual assets that align with the organization's objectives and enhance campus presence.

03. Capstone Projects

Featured Project

BU-Chatbot

Developed a RAG-based AI chatbot for the Bahria University Student Rulebook by transforming a static PDF into an interactive assistant using MongoDB Atlas Vector Search, LangChain, and Groq LLMs. Implemented conversational memory with LangChain and secure authentication with Clerk. Containerized the backend with Docker and deployed it on Azure, while hosting the frontend on Vercel, reducing response latency from 55 seconds to 5 seconds.

FastAPI
Clerk
MongoDB
Groq
LangChain
Docker
Azure

Featured Project

Anime Character Detector

Developed an end-to-end zero-shot anime character detection and tracking pipeline, covering dataset creation, model fine-tuning, optimization, and deployment. Combined a fine-tuned DEIMv2 object detector with a LoRA-adapted DINOv3 Vision Transformer for robust visual feature extraction and character re-identification without retraining for new classes. Integrated vector database-based similarity search for matching characters across images and videos, and exported the pipeline to ONNX for optimized inference. Achieved near real-time performance, processing 1 second of video content in approximately 2.5 seconds using GPU acceleration.

DEIMv2
DINOv3
Numpy
LanceDB
OpenCV
ONNX
GoogleColab

Featured Project

NYC Fare Predictor

Developed a hybrid NYC taxi fare prediction system using over 40 million trip records. Built an ensemble of five machine learning models to estimate variable fare components, combined with data-driven business rules for fixed charges derived from extensive exploratory data analysis (EDA). Achieved an average prediction error of 8.5% and deployed the application on Hugging Face Spaces with an optimized inference pipeline for fast CPU performance.

Numpy
DuckDB
SKLearn
XGBoost
Seaborn
FastAPI
HuggingFace

Featured Project

ViT Comparison

Conducted a comparative study of parameter-efficient fine-tuning techniques on DINOv3-ViT-L/16 (300M parameters) for few-shot fine-grained image classification using a custom dataset of 1,292 images across 26 classes. Benchmarked frozen retrieval, linear probing, partial fine-tuning, LoRA, and LoRA with Supervised Contrastive Learning (SupCon), where LoRA consistently outperformed full fine-tuning at every data stage (10%-80%), reaching 97.8% Top-1 accuracy. Implemented memory-efficient training with gradient checkpointing and optimized batch scheduling to overcome GPU memory limitations.