Portfolio · 2026

Adib
Sakhawat

Building intelligent systems at the intersection of software engineering and AI research — backend platforms, NLP, information retrieval, LLMs, and developer tooling.

Software Engineering AI / ML & NLP Information Retrieval Developer Tools

View my work Get in touch GitHub

SWE · NLP · IR

Location

Dhaka, Bangladesh

Status

Final year · SWE @ IUT

Open to

Remote roles

GitHub

@sakhadib

01 — About

Engineering, research, and the open web.

Adib Sakhawat is a Software Engineer and AI/ML & NLP researcher from Dhaka, Bangladesh — a final-year Software Engineering student at the Islamic University of Technology (IUT) who builds intelligent software systems, developer tools, and research-driven applications at the intersection of software engineering, NLP, Information Retrieval, and developer tooling.

I enjoy building technically ambitious systems — from production backend platforms and AI agents to programming-language tooling, code intelligence systems, and reproducible NLP research. My work often focuses on turning ideas that are difficult or unconventional into practical, working software, with a bias toward low-resource languages and culturally grounded benchmarks.

My research interests include Natural Language Processing, Information Retrieval, LLM evaluation, and computational methods for understanding and working with software — particularly language-agnostic code intelligence, semantic code retrieval, and systems that make software development more intelligent and efficient.

Alongside engineering and research, I have led technical initiatives and communities at IUT — including the IUT Computer Society and IEEE IUT Computer Chapter — and built technology platforms serving thousands of students and large-scale university events. I'm currently interested in remote opportunities in Software Engineering, AI/ML, NLP, and Information Retrieval — especially roles involving technically challenging systems, developer tools, intelligent software, or research-driven products.

Current Roles

Vice President — IUT Computer Society
Vice Chair — IEEE IUT Computer Chapter
Vice President — IUT Al-Fazari Interstellar Society
Final Year Student — Software Engineering, Dept. of CSE, IUT

02 — What I Do

Five practices.

The kinds of work I take on — spanning production engineering, applied AI, and research.

Software Engineering

Building production-ready backend systems, APIs, developer tools, and full-stack applications.

AI / ML & NLP

Working with LLMs, NLP pipelines, evaluation frameworks, generative AI, and AI-assisted systems.

Information Retrieval

Building semantic search and retrieval systems using embeddings, vector search, and language-agnostic representations.

Developer Tools

Designing tools around code intelligence, parsers, DSLs, compilers, CLI applications, and programming-language infrastructure.

Research

Developing reproducible experiments and benchmarks around NLP, LLM behavior, and intelligent software systems.

03 — Research

Areas of inquiry.

Topics I actively read, build, and write papers on.

Natural Language Processing

Information Retrieval

Large Language Models

LLM Evaluation

Code Intelligence & Retrieval

Language Tooling & Compilers

Low-Resource Language Tech

Legal AI & Computational Law

Information Extraction

ML Interpretability

Computational Mathematics

Multilingual Benchmarking

AI for Social Good

Crime Intelligence Systems

04 — Publications

Papers & preprints.

Selected work spanning ideological plasticity in LLMs, figurative language, political alignment, evaluation geometry, and game-theoretic benchmarks for dialogue.

P/01 AACL 2026 · Main Conference

LLM-Ideoplasticity: Measuring Ideological Plasticity in the Political Behavior of LLMs as a Context-Conditioned Distribution

A reproducible research pipeline studying how the political behavior of LLMs changes under different conversational and prompting conditions. Projects model responses into a shared VAA–CHES political space and measures shifts caused by prompt register, paraphrasing, reasoning, language, conversational pressure, and argumentative role.

arXiv:2606.28335 Code

P/02 LREC-COLING 2026 · Main Conference

When Words Don't Mean What They Say: Figurative Understanding in Bengali Idioms

A culturally grounded dataset of 10,361 Bengali idioms annotated with a 19-field schema capturing semantic, syntactic, cultural, and religious dimensions. Benchmarks 30 state-of-the-art multilingual / instruction-tuned LLMs — none exceed 50% accuracy versus 83.4% human.

LREC 2026 arXiv:2602.12921

P/03 ICML 2026 · MusIML Workshop · Poster

AREG: Adversarial Resource Extraction Game for Evaluating Persuasion and Resistance in LLMs

A benchmark studying persuasion and resistance as a multi-turn, zero-sum negotiation over financial resources — evaluating offensive and defensive capabilities jointly across frontier language models. The two abilities are only weakly correlated; models defend better than they attack.

arXiv:2602.16639

P/04 Dialogue Benchmark

AIDG: A Formal Decomposition of Information Extraction and Containment Asymmetries in Multi-Turn LLM Dialogue

Models adversarial information extraction and containment in multi-turn LLM dialogue as a two-player partially observable stochastic game, evaluated with Dual-ELO scoring across two tasks — social-deduction style AIDG-I and a structured "20 Questions" style AIDG-II. Across 439 games with six frontier models, containment proves far stronger than deduction (~350 ELO defensive advantage).

arXiv:2602.17443 Code

P/05 LLM Alignment Audit

Political Alignment in Large Language Models

Audits 26 contemporary LLMs using three political psychometric inventories (Political Compass, SapplyValues, 8Values) plus a large news-bias labeling task. Shows 96.3% of models cluster in the Libertarian-Left quadrant — model identity, not prompt wording, explains most variance.

arXiv:2601.06194

P/06 Evaluation Framework

Coordinates of Capability: A Unified MTMM-Geometric Framework for LLM Evaluation

A SoK paper merging multitrait–multimethod (MTMM) analysis with geometric representations to organize and interpret LLM evaluation results across many benchmarks. Treats models as points in a capability space and argues for moving from leaderboard fragments to structured, geometry-aware capability maps.

arXiv:2605.08522

Also: technical articles published in Python in Plain English on Medium.

05 — Projects

Things I've built.

Selected open-source projects across code intelligence, AI agents, language tooling, NLP, and production platforms.

PJ/01

Reko

Polyglot Code Extraction & Semantic Retrieval · Rust

A cross-platform Rust toolkit for language-agnostic code extraction and semantic code retrieval. Extracts functions and methods across 15 programming languages into a unified 215-field intermediate representation while preserving source exactly — then enables semantic search with ONNX embeddings and sqlite-vec. Parallel repo scanning, automatic model management, CI release automation.

Repository

PJ/02

Pi Daytona Chat

Sandboxed AI Agent Platform · React + Convex

A full-stack AI agent platform that runs Pi Agent inside an isolated Daytona sandbox for every conversation — separating the control plane from execution so each chat has isolated filesystem state, tools, history, and artifacts. Streaming agent events, persistent tool-call observability, and OpenRouter-based inference.

Repository

PJ/03

WebTikZ

Browser-Based TikZ Parser & Renderer

A dependency-free browser implementation of a TikZ parsing and rendering pipeline: tokenizer, recursive-descent parser, AST, geometry and expression evaluation, display-list rendering, and Canvas/SVG backends — with snapshot and visual regression testing and a PGF/TikZ corpus extractor.

Repository

PJ/04

AIManim

AI-Powered Math Animation Generator

Transforms natural-language mathematical explanations into rendered Manim animations. Automatically plans scenes, generates Manim code with LLMs, validates and repairs failed renders, generates scenes in parallel, and stitches the final video with FFmpeg. Multi-provider support: OpenRouter, OpenAI, Gemini.

Repository

PJ/05

12th National ICT Fest

Production Event Management Platform

The production platform for the 12th National ICT Fest at IUT — 30,000+ registrations across multiple competitions, with registration workflows, online payment integration, admin and faculty dashboards, reporting, and automated email / SMS / Telegram communication.

Repository

PJ/06

WTH — What The Hell

Developer Error Explainer · C++17

A cross-platform CLI that turns cryptic command-line errors into concise, actionable explanations and fixes. Regex-based error matching, automatic redaction of API keys, tokens, and passwords, and shell integration for PowerShell, Bash, and Zsh.

Repository

PJ/07

LLM-Ideoplasticity

Reproducible Research Pipeline

The pipeline behind the AACL 2026 paper — projects model responses into a shared VAA–CHES political space across multiple model snapshots, with seven analysis notebooks, transformation models, and materialized experiment artifacts.

Repo Paper

PJ/08

Minimum Line Cover Solver

Mixed-Integer Programming · C++ / HiGHS

An incremental mixed-integer programming solver built on the HiGHS optimization library — memory-efficient CSR sparse-matrix representation, incremental binary feasibility constraints, checkpoint/resume, and runtime instrumentation.

Repository

PJ/09

MathRanker

Competitive Math Platform · Laravel

A web platform for mathematical enthusiasts: contests, problem-solving, community engagement. Built with Laravel, Filament, Livewire — featuring a Sergeant-to-General rating system and collaborative forums.

Repository

PJ/10

Geon

2D Math Graphics DSL · JavaScript

A zero-dependency JS library that renders 2D mathematical graphics from a natural-language-like DSL into SVG in the browser — declarative geometry, plotting, intersection solving, smart labeling.

Repo Live Docs

PJ/11

ZotecoRD

Zotero ↔ Discord Research Bot

A Discord bot bridging Zotero with Discord and Google Sheets. Extracts color-coded PDF annotations and routes them to categorized channels — yellow → methods, green → contributions, blue → results, purple → claims, red → limitations.

Repository

PJ/12

Crime Data Scraper

AI-Powered News Pipeline · spaCy

AI-powered scraper extracting crime news from 40 verified sources across 6 continents. Uses spaCy NER and crime classification, SHA-256 + MD5 dedup, and daily GitHub Actions runs. Outputs an 18-column structured CSV.

Repository

PJ/13

CNN Comparison on Hand Gesture

Comparative CV Study · 15 Architectures

A comparative study of 15 pretrained CNN architectures for character-level static hand-gesture recognition (37 classes, 1,500 images each). Includes Grad-CAM, LIME, t-SNE, and full evaluation suites.

Repository

PJ/14

Contexto

Codebase Analysis CLI · .NET 6+

A .NET 6+ CLI that analyzes directory structures and generates XML reports about a codebase. Produces folders.xml, files.xml, stat.xml, and a combined complete.xml report — with intelligent exclusion of build folders.

Repository

PJ/15

MathVoyage (vmath)

Math Toolkit · Java

A Java toolkit covering algebra, trigonometry, matrices, vectors, number systems, combinatorics, coordinate geometry, and bitwise ops — with JavaDoc and a JAR release.

Repository

Plus roughly 130 public repositories on GitHub covering software engineering, AI/ML, NLP, developer tooling, programming languages, and research.

06 — Datasets

Curated datasets.

Public datasets I've built and released on Kaggle to support research in low-resource NLP, legal AI, and game simulation.

DS/01 10,361 entries

Bangla Bagdhara

The largest computational resource for Bengali idioms — 19-field schema, 37,129 semantic tag assignments, sentiment & cultural metadata. Accompanies the LREC-COLING 2026 paper on figurative understanding.

DS/02 1,484+ acts

Bangladesh Legal Acts Dataset

Bangladesh's legal framework (1799–2025) — 35,633 sections, 14,523 footnotes, government & legal-system context. Multilingual: English, Bengali, mixed. CC-BY 4.0.

DS/03 100,000 tournaments

Snake & Ladder Game Simulation

100,000 simulated tournaments across 16 psychologically-inspired AI agents — group stage, knockouts, and full replay sequences. ~2.3GB CSV for behavioral modeling and strategy analysis.

DS/04 7,117 headlines

Crime Headline Binary Classification

A perfectly balanced dataset (3,562 crime / 3,562 non-crime) for binary classification — useful for logistic regression, decision trees, and deep-learning baselines.

DS/05 2004 – 2024

Bangladesh Math Olympiad Question Set

Two decades of BdMO papers, organized by year, level (Primary → Higher Secondary), and round (Divisional, Regional, National, Selection).

07 — Engineering

Stack & practice.

Languages

Rust Python C++ TypeScript / JavaScript PHP SQL C# Java

AI / ML / NLP

Large Language Models NLP pipelines LLM evaluation Generative AI Agentic systems Machine learning

Information Retrieval

Semantic search Embeddings Vector search ONNX inference Code intelligence

Backend, Tooling & Infrastructure

REST APIs Laravel Database design MySQL / SQLite Microservices Parser design DSLs & AST tooling CLI development CI/CD & GitHub Actions Azure Cloud sandboxing Playwright & E2E testing Linux / WSL2

GitHub

130+

Public Repos

4+

Years on GitHub

250+

Peak Commits / Quarter

12

Languages Used

Most active languages

JavaScript / TypeScript01
Python02
Rust03
C++04
PHP / Blade05
C# / Java06

Top repositories by stars

IUT_SWE_QuestionBank01
Reko02
Ground-News-Scraper03
AiManim04
auto-indent05
MathRanker06

08 — Experience

Work & leadership.

Dec 2025 — Sep 2026

Vice President

IUT Computer Society
- Built and maintained the 12th National ICT Fest platform supporting 30,000+ registrations.
- Led technical delivery for ICT Fest, CodeSprint, and DevSprint.
- Convener and judge for CodeSprint 2.0 and DevSprint 1.0.
- Organized RISC-V Day, bringing semiconductor and hardware research sessions to campus.
Sep 2025 — Apr 2026

Software Engineer Intern

mAdEstic Software Solutions Inc.
- Worked on production ERP systems serving two major brands.
- Designed database architectures with focus on data modeling, entity relationships, and scalability.
- Built and maintained production backend APIs using Laravel and TypeScript.
- Worked with CI/CD pipelines, Azure, and GitHub Actions; automated end-to-end testing with Playwright.
- Used AI-assisted and agentic development workflows for implementation and debugging.
2025 — Present

Vice Chair

IEEE IUT Computer Chapter
Dec 2025 — Sep 2026

Vice President

IUT Al-Fazari Interstellar Society
Oct 2024 — Dec 2025

Assistant Director, Technical Affairs

IUT Computer Society
- Organized CodeRush 2.0 with approximately 1,500 registrations and built its registration and participant-management website.
- Contributed to GameJam and Datathon problem setting and contest management for the 11th National ICT Fest.
Sep 2024 — Dec 2025

Joint Secretary

IUT Al-Fazari Interstellar Society
2023 — 2024

Junior AI/ML Executive

IUT Computer Society
2023 — 2024

Junior Executive, Administration

IUT Al-Fazari Interstellar Society
2020 — 2022

Editor

Bigganil — A Bangla Science Magazine

09 — Education

Education & milestones.

My academic path, and a few markers along the way.

Undergraduate

B.Sc. in Software Engineering

Islamic University of Technology (IUT)

Aug 2022 — Oct 2026 (expected)

Higher Secondary

HSC, Science

New Govt Degree College, Rajshahi

Secondary

SSC, Science

Govt. Laboratory High School, Rajshahi

Achievements

Research accepted at the AACL Main Conference.

Research accepted at ICML 2026 and EMNLP 2026 workshops.

Paper presented at the LREC-COLING 2026 Main Conference.

Built a production event platform handling 30,000+ registrations.

Developed a polyglot code extraction & semantic retrieval toolkit supporting 15 programming languages.

Maintained approximately 130 public GitHub repositories.

Student volunteer at ICML 2026, Seoul.

10 — Beyond

Outside the terminal.

Astronomy & space science

Stars, cosmology, observational astronomy — and the societies that gather around them.

Science fiction

Sci-fi and science-themed media — narrative as a way to think about systems.

Ancient Bengal history

Civilizations, manuscripts, and the long arc of cultural memory in the subcontinent.

Research discussions

Long-form academic conversations — the kind that shape future papers.

11 — Vision

What I'm building toward.

To build impactful AI systems, contribute meaningful research to the NLP community, and develop technologies that help low-resource languages and human ecosystems gain stronger representation in modern AI — while making software development itself more intelligent and efficient.

12 — FAQ

Questions, answered.

The short version of who I am, what I work on, and how to reach me.

Who is Adib Sakhawat?

Adib Sakhawat is a Software Engineer and AI/ML & NLP researcher from Dhaka, Bangladesh — a final-year Software Engineering student at the Islamic University of Technology (IUT) who builds intelligent software systems, developer tools, and research-driven applications, and publishes on LLM behavior and low-resource language technology.

What does Adib Sakhawat work on?

Software engineering, AI/ML and NLP, information retrieval, and developer tooling: production backend platforms, semantic code retrieval, polyglot code extraction, LLM evaluation, AI agents, parsers and DSLs, and reproducible NLP research.

Where has Adib Sakhawat published research?

At the AACL 2026 Main Conference (LLM-Ideoplasticity), the LREC-COLING 2026 Main Conference (figurative understanding in Bengali idioms), the MusIML Workshop at ICML 2026 (AREG), and an EMNLP 2026 workshop — alongside preprints on political alignment in LLMs, dialogue information asymmetry (AIDG), and geometric LLM evaluation frameworks.

Is Adib Sakhawat open to remote work or research collaboration?

Yes — open to remote opportunities in Software Engineering, Backend Engineering, AI/ML, NLP, Information Retrieval, Developer Tools, and Research Engineering, as well as research collaboration on NLP, legal AI, and low-resource language technology.

How can I contact Adib Sakhawat?

By email at sakhadib@gmail.com, by phone or WhatsApp at +880 1645-534121, or via GitHub and LinkedIn.

13 — Contact

Let's build something interesting.

I'm open to remote opportunities in Software Engineering, Backend Engineering, AI/ML, NLP, Information Retrieval, Developer Tools, and Research Engineering — as well as research collaboration and conversations on legal AI and low-resource language technology.

sakhadib@gmail.com

+880 1645-534121

Phone / WhatsApp