Computer engineering graduate interested in algorithms, software and artificial intelligence.
Looking for a master's place in the computer field
Pick a benchmark instance and watch HBALS solve it live, or tap the map to add your own cities. It is a JavaScript port of the algorithm in my repository, with the settings of the final version (one swap per candidate, no adaptive neighbourhood). To finish quickly, the demo stops after 20 iterations without improvement, whereas the study used a fixed budget. How it works. Instances are from TSPLIB (Reinelt, 1991).
A hybrid Bees Algorithm with 2-opt local search, which I designed for the Traveling Salesman Problem. It was my bachelor's final project, and I am the sole author. The revised study, with an equal-budget comparison and an ablation, is on Zenodo as a preprint (not yet peer reviewed), and I aim to get it published in a peer-reviewed journal.
Given a set of cities, find the shortest route that visits each one once and returns to the start. The problem is NP-hard, so for larger inputs the practical goal is a very good route found quickly, not a proof that it is the best.
The standard Bees Algorithm can converge slowly or stall in local optima. Hybrids with local search are common, but they are often compared with weaker baselines. My question became: what does a Bees-style hybrid contribute once every method gets the same local search and the same computing budget, and which of its parts actually matter?
In the algorithm itself, each bee is a candidate route. Bees search close to the best routes they know, while scouts try random new ones. The animation runs the real search in the background and uses the yellow dots to let you follow the best route.
| TSPLIB instance | Cities | Best known | HBALS best found | HBALS mean gap (%) | ILS-2opt mean gap (%) |
|---|---|---|---|---|---|
| eil51 | 51 | 426 | 426 ✓ | 0.035 | 0.200 |
| berlin52 | 52 | 7,542 | 7,542 ✓ | 0.000 | 0.154 |
| st70 | 70 | 675 | 675 ✓ | 0.030 | 0.348 |
| kroA100 | 100 | 21,282 | 21,282 ✓ | 0.008 | 0.117 |
| ch150 | 150 | 6,528 | 6,549 | 0.545 | 0.541 |
| rat195 | 195 | 2,323 | 2,338 | 1.184 | 0.837 |
20 independent runs per method, the same budget of move evaluations for every method (equivalent to 300 HBALS iterations) and the same 2-opt operator. The comparison also includes a memetic genetic algorithm, a memetic ant colony, multi-start 2-opt and a plain Bees Algorithm. Gaps are means over the 20 runs. A ✓ marks a best tour equal to the best-known one. The six instances shown are the ones you can run live; the study used 12 main instances and three larger ones.
Over 12 instances with 51 to 198 cities, a Friedman test shows that the methods differ (p < 0.001). HBALS is significantly better than the genetic, ant colony, multi-start and plain Bees Algorithm methods (Holm-adjusted p = 0.0024), but not significantly different from iterated local search overall (p = 0.13). On the larger instances a280 and lin318, iterated local search is significantly better. An ablation showed that 2-opt at the elite sites and a single-swap perturbation drive the performance. I therefore do not claim a new state-of-the-art solver. My first version compared HBALS with baselines that had no local search; the revised study corrects this, and the repository lists the corrections. The full implementation, data and results are open source under the MIT licence.
A 12-slide presentation for my Artificial Intelligence and Expert Systems course (2025). It explains how real honey bees forage, how the Bees Algorithm turns that into a search method with its parameters, pseudocode and flowchart, and where it is used. The Bees Algorithm is the method HBALS builds on.
A 19-reference report on Light Fidelity for the Research Methods and Presentation course. It covers modulation methods (OOK, VPPM, OFDM, CSK) and uses in healthcare, aviation and underwater communication.
I am a Computer Engineering graduate with a strong foundation in algorithms, programming, computer networks and artificial intelligence.
I designed HBALS, build web apps with React and JavaScript, and like problems where there is a clear way to measure whether the answer got better.
I am looking for a master's place in the computer field, building on my major.
Code is public on GitHub. Demos run in your browser.
A browser-based resume builder in vanilla JavaScript, HTML and CSS, with five customizable templates, a live preview and export to PDF, Word, Excel and text. The AI Job Match tool takes a pasted job description and returns a match score, missing skills and tailoring suggestions.
A multi-page online-shop front end in React with client-side routing, product pages, search and a cart saved in the browser. It works in Persian and English, right-to-left and left-to-right. The interface is inspired by an existing Iranian shop and was built for learning.
All twelve projects from Harvard's CS50AI, from classical AI to neural networks: search (BFS, minimax with alpha-beta pruning), logic (Knights and Knaves, Minesweeper), probability (PageRank, Bayesian networks), a crossword solver with AC-3 and backtracking, k-NN and Q-learning, a CNN for traffic signs, CFG parsing and BERT attention analysis.
Four small applications: a library management system with JSON storage, a password manager, a software company system in object-oriented design, and a NumPy analysis of monthly sales.
Islamic Azad University, Ardabil Branch, Iran · September 2022 – July 2026
The university certified a GPA of 18.05 out of 20 over 152 credits and a rank of 4 among the graduates of the second semester of the 2025–2026 academic year. It was issued on 13 October 2026; this is the certified English translation of the Persian original.
Personal numbers and QR codes are blurred for privacy.
I sat the exam at the British Council test centre in Van, Turkey, on 4 September 2026.
CS50
CS50
Circuits lab
Circuits lab
Before this internship, networks were something I drew on paper and configured in lab exercises. At the port in Astara they were cabinets, cables and people who needed everything to just work.
I got to work on real equipment: switches and patch panels, VLANs, VPN, DNS and DHCP. I learned that troubleshooting is mostly patience and knowing the next good question to ask, and I enjoyed it far more than I expected.
Working in a port was a special experience. It is a place that keeps moving, and the network quietly holds a lot of that together. It made me care about systems that people depend on without ever thinking about them.
Based in Astara, Iran.