ML / AI Engineer · Researcher

Building reliable AI systems for real‑world impact.

I specialize in LLMs, RAG systems, and computer vision, grounded in a strong statistics foundation. I build clean, scalable, production-ready AI systems and research tools that deliver real impact

View projects ↓ Hire / collaborate Download resume

Noakhali, Bangladesh · B.Sc. in Statistics (NSTU, 2022–2026)

Phone: 01730644634

Email: azizashfak@gmail.com

GitHub ↗ LinkedIn ↗ Kaggle ↗ LeetCode ↗

Aziz Ashfak

ML/AI Engineer · LLMs · RAG · CV

Open to roles & collaborations

Experience

AI Engineer · ROYALX LLC

Remote · Dhaka, Bangladesh · Aug 2025 – Nov 2025 · Part-time

LLM / RAG

Built LLM apps (OpenAI, Groq, HF) with LangChain/LlamaIndex for generation, summarization, and chatbots.
Designed multimodal RAG pipelines to handle text + figures with improved retrieval accuracy.
Fine‑tuned transformers for QA, classification, and document understanding tasks.
Developed agents for reasoning‑heavy, multi‑step workflows and automation.
Deployed production APIs/UIs with FastAPI, Flask, Streamlit, and CI/CD.

AI Tutor · Ai Devlop

Remote · Dhaka, Bangladesh · Apr 2025 – Aug 2025 · Part-time

Teaching

Taught 100+ students statistics, EDA, ML, and deep learning through practical labs.
Created structured curricula, capstones, and personalized mentoring.
Guided students to build and deploy end‑to‑end ML solutions.

Featured projects

Each card includes code, live app, and a demo video.

Regression · Flight Ticket

Machine Learning

Indian Flight Price Prediction

BaggingRegressor · 300k+ rows

High‑accuracy price model (≈ 99.7%) with feature engineering, preprocessing, and CI/CD deployment.

scikit‑learn Pandas Numpy Matplotlib Seaborn Plotly Flask Git Github Action Render Moduler Coding

Code ↗ Live ↗ ▶ Demo

Classification · Healthcare

Machine Learning

Thyroid Disease Prediction

RandomForest · Render

≈ 98.7% accuracy with tuned hyperparameters and an end‑to‑end inference pipeline.

scikit‑learn Pandas Numpy Matplotlib Seaborn Plotly Flask Git Github Action Render Moduler Coding

Code ↗ Live ↗ ▶ Demo

Computer Vision

Deep Learning

Brain Tumor Detection

ResNet152V2 / DenseNet121

≈ 98.5% accuracy on MRI images; exported to ONNX for fast web inference.

TensorFlow Keras Transfer Learning MLFlow ONNX Flask Git Github Action Render Moduler Coding

Code ↗ Live ↗ ▶ Demo

GenAI · Multimodal

Multimodal RAG

Multimodal Research Paper Analyst

LLM · Streamlit

Reads PDFs & figures, summarizes sections, and exports to Word/PDF/CSV/Markdown.

LangChain Groq Llama OpenAI Transformer FAISS Hugging-Face Streamlit Moduler Coding

Code ↗ Live ↗ ▶ Demo

GenAI · RAG

RAG

QnA RAG Bot for PDFs

FastAPI · LangChain

Figure‑aware retrieval, modular pipelines, multi‑format export for research workflows.

LangChain Groq Llama OpenAI Transformer FAISS ChromaDB Hugging-Face FastAPI Moduler Coding

Code ↗ Live ↗ ▶ Demo

Computer Vision · Real‑time

Object Detection

Face Mask Detection

ONNX · Flask

Real‑time detector on a Kaggle dataset using YOLO; ONNX export, served via Flask app.

TensorFlow Keras YOLO Pandas Numpy MLFlow ONNX Flask Git Github Action Render Moduler Coding

Code ↗ Live ↗ ▶ Demo

Research focus

U‑Net for Human Segmentation

Encoder–decoder optimization · ICRA 2026 (proposed)

Optimizing U‑Net architectures for human segmentation enhances accuracy and efficiency in computer vision tasks. This research advances robust models for healthcare, surveillance, and real‑world applications.

Benchmarking Large Language Models on Bangla to advance low‑resource multilingual NLP evaluation.

GenAI · LLMs

Benchmarking Large Language Models on Bangla highlights the challenges of low‑resource multilingual NLP. This work advances inclusive evaluation frameworks to improve performance and accessibility for underrepresented languages.

Technical stack

Core ML / AI

Languages & ML/DL

Python SQL R PyTorch TensorFlow Keras scikit‑learn OpenCV MLflow C(basic) Java(basic)

GenAI · LLMs · RAG

LLM & retrieval stack

OpenAI API Hugging Face Groq LangChain LlamaIndex LangGraph CrewAI n8n FAISS ChromaDB Pinecone

Deployment · Data

MLOps & analytics

FastAPI Flask Streamlit Gradio ONNX Git GitHub Actions Render Kaggle Google Colab

Pandas NumPy Matplotlib Seaborn Plotly

DSA & Problem Solving

Milestone Achieved

Completed most core Data Structures & Algorithms with hands‑on practice and problem‑solving rigor.

Arrays Strings Hash Maps Linked Lists Trees Graphs Recursion Sorting/Search Dynamic Programming

LeetCode: leetcode.com/azizashfak

Achievements

Teaching & Mentoring

100+ students · Capstone guidance

Project‑based curricula; mentored cohorts to ship apps (ML APIs, dashboards, research notebooks).

Open‑source & Community

Reusable notebooks · Annotated guides

Published guides across ML, CV, and RAG, focused on reproducibility and explainability.

Testimonials

“Aziz’s mentoring transformed our cohort’s confidence. His annotated notebooks made complex ML concepts approachable.”

— Student, Ai Devlop

“Production‑first thinking. He ships reliable pipelines and explains tradeoffs with refreshing clarity.”

— Collaborator, ROYALX LLC

Blog & notes

End to End ML Notebook

scikit‑learn · Data Analytics

Developed an end‑to‑end Flight Price Prediction pipeline on Kaggle, covering data preprocessing, feature engineering, model training, and evaluation. Delivered a reproducible notebook with clear insights and deployment‑ready predictions for real‑world use cases.

End to End DL Notebook

TensorFlow · Transfer Learning

Developed an end‑to‑end Brain Tumor Detection pipeline on Kaggle using CNNs and transfer learning. The notebook covers preprocessing, model training, evaluation, and explainable visualizations to enhance clinical interpretability.

EDA

Pandas · NumPy · Data visualizations

Performed exploratory data analysis on Netflix titles, uncovering trends in genres, release years, and content distribution. Delivered a reproducible notebook with clear visualizations and insights into platform growth and audience patterns.

Future work interests

Explainable AI in healthcare

Grad‑CAM · SHAP · Clinical trust

Interpretable diagnostic models for cancer and brain imaging to enhance clinician confidence and decision support.

Multilingual NLP for Bangla

Benchmarking · Data curation · Evaluation

Robust benchmarks and datasets for underrepresented languages focusing on retrieval, reasoning, and fairness.

Future project ideas

Idea · Not started yet

Agentic RAG Evaluator

Stress‑test RAG systems with adversarial questions, hallucination checks, and figure‑aware retrieval metrics.

Planned stack: Llama 3, LangGraph, FastAPI, FAISS, Weaviate, Postgres, LangChain, OpenAI API, Hugging Face Transformers, Docker, GitHub Actions, Streamlit, React, Tailwind CSS.

Idea · Not started yet

Intelligent Video Calling Agent

Build a web-based agent that enables real-time video calling with integrated AI features like transcription, summarization, emotion detection, or even retrieval-augmented responses.

Stack : WebRTC, MediaPipe,,Whisper, LangChain, Hugging Face Transformers, FAISS, GPT-4,FastAPI,WebSockets, Redis, Postgres, Deepgram, OpenAI API

Contact

Prefer email?

Email Aziz

Phone: 01730644634 · Noakhali, Bangladesh

Open to roles & collaborations in LLMs, RAG, and computer vision.

Explore projects Download resume