I
Software Engineering for AI (SE4AI)
Bringing engineering rigor to AI systems: reliable pipelines, testability, and the design practices that make ML-powered software dependable.
Researcher — lens 02Research Assistant
The space where software engineering meets machine intelligence.
Research statement — I study the boundary between software engineering and machine intelligence — how intelligent systems can assist the engineering process, and how engineering discipline can make intelligent systems more reliable.
Research
Applying machine learning and LLMs to assist software development: automated testing, code assistance, and intelligence inside engineering workflows.
II
AI for Software Engineering (AI4SE)
Applying machine learning and LLMs to assist software development: automated testing, code assistance, and intelligence inside engineering workflows.
III
Security in Software Systems & LLMs
Studying the security surface of software systems and of large language models used inside them — from intrusion detection to model-assisted attacks.
IV
Metaheuristic Optimization & Feature Selection
Nature-inspired search methods (e.g. slime mould optimization) applied to high-dimensional feature spaces, plus resource allocation and scheduling in cloud–edge systems.
My research questions come from software I have actually shipped. That grounding — knowing how systems break in practice — is the lens I bring to the lab.
Interests
These are the interests I actively pursue. Any topic outside this list has not yet earned my conviction — and will appear here honestly only when it does.
Timeline
Milestones are recorded as they happen. Entries marked as pending reflect work in progress, never claimed results.
NowIslamic Azad University, Central Tehran Branch
M.Sc. in Software Engineering
Graduate studies focused on AI × software engineering, alongside industry engineering work at WebCore.
2025The Journal of Supercomputing
First journal publication
Multivariate filter feature selection with stacking-based ensemble learning for network intrusion detection — published after peer review.
2025Industry
Software Engineer at WebCore
Converted from a 6-month internship into a full-time frontend engineering role based on performance.
2023 — 2025Two-person programming team
Freelance backend development
Delivered client systems — including a CI/CD automation tool in Go — as the backend half of a freelance team.
2019 — 2024Rouzbahan University
B.Sc. in Biomedical Engineering
Graduated with a 17/20 GPA; thesis applied PCA and CNNs to coronary artery disease detection from medical imaging.
Publications
- [1]2025The Journal of SupercomputingJournal article
Multivariate Filter Feature Selection with Stacking-Based Ensemble Learning for Network Intrusion Detection
Hashemi Jouybari, S. M., Janbaz, A., & Esfandiari, A.
Research projects
Investigations structured as research; each entry follows the same five-part anatomy: question, method, experiment, result, contribution.
Preprint
01
MetaheuristicsFeature selection
A Redundancy-Aware Hybrid Feature Selection Framework Using Mutual Information Clustering and Enhanced Slime Mould Optimization
Research Question
Can pairing mutual-information clustering with an enhanced slime mould optimization algorithm remove feature redundancy more effectively than filter or wrapper methods alone?
Method
Hybrid framework: mutual information clusters correlated features; an enhanced slime mould optimization algorithm searches within and across clusters for compact, high-value subsets.
Experiment
Benchmarked against standard filter and metaheuristic baselines on classification datasets.
Result
Manuscript in submission.
Contribution
Co-developed the framework and co-authored the manuscript.
02
Feature selectionStacking ensembles
Multivariate Filter Feature Selection with Stacking-Based Ensemble Learning for Network Intrusion Detection
Research Question
Does combining multivariate filter feature selection with a stacking ensemble improve intrusion-detection performance over single models?
Method
Multivariate filter selection reduces the feature space; a stacking architecture combines base classifiers for the final decision.
Experiment
Evaluated on network-intrusion benchmark data with standard detection metrics.
Result
Published as a peer-reviewed journal article in The Journal of Supercomputing (2025).
Contribution
Co-developed the selection method and stacking architecture.
03
Cloud–edge computingScheduling
Recent Advances in Resource Allocation and Task Scheduling for Heterogeneous Cloud–Edge Systems
Research Question
What strategies exist for allocating resources and scheduling tasks across heterogeneous cloud–edge infrastructures — and where are the open gaps?
Method
Systematic review and categorization of recent resource-allocation and task-scheduling algorithms.
Experiment
Coursework survey paper (Advanced OS, M.Sc.).
Result
Identified open challenges across latency-aware, energy-aware, and QoS-aware scheduling strategies.
Contribution
Authored the survey as M.Sc. coursework.
04
Requirements engineeringSurvey
Requirement Elicitation and Analysis Techniques in Modern Software Engineering
Research Question
Which requirement-elicitation and analysis techniques dominate modern software engineering practice?
Method
Review of elicitation and analysis techniques (Advanced Software Engineering, M.Sc.).
Experiment
Coursework article.
Result
Coursework article completed (2025/2026).
Contribution
Authored the review as M.Sc. coursework.
05
Deep learningMedical imaging
Coronary Artery Disease Detection using PCA and CNNs
Research Question
Can a CNN pipeline detect coronary artery disease from medical imaging data reliably enough for screening support?
Method
PCA for feature selection and dimensionality reduction, followed by a convolutional neural network for classification.
Experiment
Trained and evaluated on medical imaging data.
Result
Achieved 90% classification accuracy.
Contribution
Designed and implemented the full pipeline (B.Sc. thesis, Rouzbahan University, 2024).
06
XGBoostAudio features
Hierarchical Music Genre Classification Using XGBoost and Audio Feature Analysis
Research Question
How well can a hierarchical genre-mapping strategy with gradient-boosted trees classify music from audio features alone?
Method
ML pipeline over 114,000 Spotify tracks and 15 audio/acoustic features; hierarchical genre mapping consolidates fine-grained genres into six top-level categories (Pop, Rock/Metal, Electronic, Latin, Hip-Hop/R&B, Classical).
Experiment
Trained an XGBoost multiclass classifier with stratified train/test splitting and tuned hyperparameters.
Result
75% test accuracy and a 0.75 weighted F1-score.
Contribution
Built the end-to-end pipeline and evaluation (2025).
Education
Current
M.Sc. in Software Engineering
Islamic Azad University, Central Tehran Branch · Sep 2025 — Present
Current GPA: 17.42 / 20.00. Graduate research at the intersection of software engineering and machine learning.
Relevant coursework
- Advanced OS
- Advanced Software Engineering
Achievements
1st Place — Quera Golang HamCode Competition (Individual) · 3rd Place — Quera CodeCup Competition (Team)
Research direction
Current direction
Working at the intersection of software engineering and machine intelligence: feature selection and ensemble methods for security, metaheuristic optimization, and how engineering discipline can make ML systems dependable.
Future interests
- 01Security in software systems and large language models
- 02Resource allocation and task scheduling in cloud–edge systems
- 03Applied machine learning and ensemble learning
- 04Engineering practices for reliable AI-powered software
For prospective supervisors: I'm looking for environments where engineering discipline and research rigor reinforce each other — projects at the intersection of AI and software engineering are the strongest fit.
Open to opportunities
Contact
Open to research assistantships and PhD opportunities in AI4SE, SE4AI, LLMs, and AI software testing.