Applied AI engineering and experimental machine learning.
My work combines applied AI engineering with experimental machine learning. During my internship at EkkBaz, I developed multi-agent workflows and retrieval systems for B2B supply chains. My research projects cover few-shot question answering, radio-signal classification, and time-series forecasting.
I co-authored an IEEE conference paper on black-hole detection using machine learning. I’m now extending that foundation through postgraduate study at Deakin University, with an interest in AI systems supported by careful data preparation, evaluation, and clear communication of results.
Engineering projects
AI systems, data pipelines, and collaborative builds.
Built a three-agent workflow for policy-document review using Google ADK, Gemini, and MCP. Integrated document access and notification tools, with a containerised FastAPI service deployed to Cloud Run.
Google ADK · Gemini 2.5 · MCP · Cloud Run
Technical details
Orchestrated Monitor, Authoriser, and Comparison agents. Integrated Google Drive and Gmail tools, authentication, and error handling, with a Docker-containerised FastAPI/Uvicorn service deployed to Google Cloud Run.
Developed a processing and regression pipeline for more than 3.7 million NYC TLC trip records, covering Parquet ingestion, data validation, feature engineering, and model comparison.
ETL · Feature engineering · Regression
Technical details
Preprocessing and validation supported comparisons of linear regression and random forest regression. Project information is available on my LinkedIn profile.
Integrated multiple NHANES survey cycles into a healthcare analytics pipeline, with automated file conversion, missing-value handling, feature engineering, and interactive Bokeh visualisations.
Data wrangling · ETL · Bokeh
Technical details
Automated XPT-to-CSV conversion, data integration, feature engineering, and missing-value handling. Project information is available on my LinkedIn profile.
Software engineering · UNIHACK 2026 · Team project
Co-built a campus food-sharing marketplace with separate cook and customer views, real-time Firestore updates, ordering flows, and university-email verification.
React · TypeScript · Firebase · Team project
Technical details
Built at UNIHACK, held at Monash University’s Clayton campus. The stack used Vite, Tailwind, Firebase Auth and Firestore, with role-based access, Google OAuth, pricing logic, and trust-and-safety controls.
Contributed to a framework combining dynamic attention, a simplified mixture of experts, and learnable positional encoding. Work included tokenisation, span alignment, and fine-tuning transformer models with limited training samples.
Transformers · Few-shot QA · Mixed precision
Technical details
January–December 2024. Work included a 10k-vocabulary BPE tokenizer, span alignment, adaptive masking, and fine-tuning BERT-family models on limited samples using cosine annealing, gradient clipping, and early stopping.
Developed a three-branch CNN ensemble to classify seven types of radio-signal spectrograms. The workflow included class balancing, leakage checks, multi-seed evaluation, bootstrap confidence intervals, and Grad-CAM interpretation.
Approximately 95% classification accuracy reported in the project experiments.
TensorFlow · CNN ensembles · Grad-CAM
Technical details
June 2023–April 2024. Combined InceptionResNetV2, ResNet50, and MobileNetV2 with SpecAugment, class balancing, MD5 leakage checks, multi-seed evaluation, bootstrap confidence intervals, and Grad-CAM interpretation.
Compared LSTM, XGBoost, and ARIMA using historical Alpha Vantage data, with stationarity checks, seasonal decomposition, and feature engineering.
LSTM achieved a reported MAPE of 1.32% in the project evaluation.
Python · LSTM · XGBoost · ARIMA
Technical details
January–April 2025. Preprocessing, stationarity checks, seasonal decomposition, feature engineering, and model evaluation. The reported error is specific to this experiment; it is not a guarantee of future forecasting performance.
Reported model results relate to each project’s experimental setup; they do not imply performance on new datasets or future market data.
Experience
Professional work and practical contributions.
Artificial Intelligence Engineer Intern
September 2024 – March 2025
EkkBaz Bangladesh Pvt. Ltd. · Dhaka, Bangladesh
Developed a LangChain and LangGraph multi-agent system for a B2B supply chain, deployed through Azure AI Studio.
Integrated real-time updates with Django REST APIs and coordinated planning, execution, and verification agents.
Reported outcomes included approximately 35% less vendor-matching time and an 18% improvement in retrieval accuracy through RAG and prompt tuning.
Additional experience in Australia
Hotel Room Attendant · Part-time
Meriton Suites Melbourne · July 2026 – present
Alongside postgraduate study, maintain guest rooms to five-star standards and coordinate room readiness with front office teams.
Education & skills
Academic background and technical toolkit.
Master of Data Science (Professional)
Deakin University · Victoria, Australia
February 2026 – November 2027 · Expected completion
Deakin International Meritorious Scholarship · 25%
B.Sc. in Computer Science and Engineering
North South University · Dhaka, Bangladesh
January 2019 – April 2025
Skills & tools
AI systems
LangChain · LangGraph · Google ADK · RAG · MCP
Machine learning
PyTorch · TensorFlow · Scikit-learn · Hugging Face Transformers
Roles, research, and opportunities to collaborate.
I welcome conversations about AI engineering and data science roles, research collaborations, and projects that connect machine learning with practical problems.
