Azizul Haque
Computer Vision Kernel

Original Image

No kernel applied. The raw RGB image data.

Open to opportunities

Machine Learning
Engineer & AI Researcher

Building large language model evaluation systems, real-time accessibility pipelines, and scalable ML solutions. AI Research Assistant at Stevens Institute of Technology · MS in Applied Artificial Intelligence.

Building AI at Scale

Hi, I'm Azizul. My interests lie at the intersection of AI research and engineering. I've built and deployed AI models in production for trillion-dollar companies including Aramco, bringing cutting-edge machine learning solutions to enterprise scale.

Areas of Focus

Large Language Models (LLMs) LLM Truthfulness Evaluation Transformer Fine-Tuning Natural Language Processing Mental Health NLP Accessibility-Focused AI Real-Time AI Systems Delay-Buffered Editing Systems Human-Centered AI LLM Benchmarking (TruthfulQA) Log-Likelihood Evaluation Batch GPU Inference Hugging Face Transformers TensorFlow & PyTorch Perceptual Hashing (pHash) Video & Audio Processing (ffmpeg, moviepy) LLM Hallucination Analysis Scalable ML Experimentation

Career Journey

AI Research Assistant

Researching large language models (LLMs) and real-time accessibility systems at Stevens Institute of Technology. Built automated truthfulness evaluation pipelines using TruthfulQA and developed delay-buffered editing systems for accessible slide presentations (ACM ASSETS 2025).

Data Analyst

Analyzed large-scale datasets at NEXT Ventures to drive business decisions. Built SQL/Python-based analytics pipelines, dashboards, and statistical models to improve operational efficiency and data-driven strategy.

Senior Web Developer

Led full-stack development projects at ExertPro LLC, building 30+ scalable web applications using React, JavaScript, and PHP. Integrated AI-driven features and improved performance, reducing load times by 60%.

Frameworks & Tools

Machine Learning & NLP

Python PyTorch TensorFlow Scikit-learn Pandas NumPy NLTK spaCy Matplotlib

LLM & Transformer Systems

Hugging Face Transformers GPT-2 GPT-Neo DistilBERT MentalBERT TruthfulQA Benchmark Log-likelihood Scoring Model Fine-tuning

Research & Experimentation

Jupyter Notebook Google Colab GPU Inference Batch Evaluation Pipelines Human-Subject Study Design ACM Publishing LaTeX Overleaf

Accessibility & Media Processing

pydub moviepy ffmpeg Perceptual Hashing (pHash) Real-time Editing Pipelines Silence Detection

Data & Analytics

SQL Data Visualization Statistical Analysis Dashboard Development Large-scale Dataset Processing

Web & Application Development

React JavaScript PHP HTML CSS Flask FastAPI Full-Stack Development

Systems & Tools

Git Linux Docker TensorBoard

Publications & Education

Accessible Slide Presentation via Intelligent Real-time Editing

Published in ACM ASSETS 2025. Insights from a Wizard-of-Oz Study on real-time accessibility editing.

Delay-Buffered Editing for Accessible Slide Presentation Videos

Under review at ACM Designing Interactive Systems (DIS 2026).

ASPIRE: Accessible Slide Presentation via Intelligent Real-time Editing

Under review at ACM CHI 2026.

🎓 M.Eng in Applied Artificial Intelligence, Stevens Institute of Technology (Expected Dec 2026)
🎓 B.Sc in Computer Science & Engineering, City University (2016–2020)

AI Research & Systems Engineering Specializing in LLM evaluation, accessibility-focused AI systems, and real-time ML pipelines.
View Publications on LinkedIn

AI Playground

Interactive visualizations of core ML/AI concepts

Neural Network Visualizer

Watch forward & backward propagation through a neural network. Hover nodes to see activation/gradient values.

Click input nodes for forward pass, output nodes for backprop

Vector Embedding Space

Explore word embeddings in 2D space. Drag to pan, scroll to zoom.

Similar words cluster together in embedding space

Self-Attention Heatmap

Visualize transformer self-attention patterns between tokens.

Hover over cells to see attention weights

PyTorch AutoGrad: Gradient Descent

Watch how different optimizers navigate a 2D loss surface. Click to place starting point.

0.020
SGD
Momentum
Adam
Gradient

Click anywhere on the surface to start optimization. "Compare All" runs all optimizers simultaneously.

AI Inspiration 🎨 Text→Image

Multimodal API + Prompt Engineering

Images generated via Pollinations.ai Text-to-Image API with crafted prompts.

Each affirmation uses prompt engineering techniques—specific keywords, style modifiers, and seeds—to generate consistent, themed AI artwork.

Click for more • Powered by Pollinations.ai
Azizul
Chat with Azizul Grounded LLM via Gemini + Cloudflare
⚡ Async

⚡ Async/Await Pattern

This chatbot uses modern JavaScript async/await for non-blocking API calls to the Gemini LLM backend.

const response = await fetch(api);
Hi! 👋 I'm Azizul's AI assistant powered by Gemini with grounded responses. Ask me anything about his experience, skills, or research!