I study how large language models behave when we measure them: how their behavior shifts with context, how they persuade and resist, and how well they understand languages the field often leaves behind.
Adib Sakhawat is a Bangladeshi NLP researcher and software engineer, and an Affiliate Researcher at the Systems and Software Lab, Islamic University of Technology, where he completed a BSc in Software Engineering with First Class Honours in 2026. His research asks how large language models behave when we measure them: how their positions shift with context, how they persuade and resist, and how well they understand languages such as Bangla that mainstream benchmarks leave behind. His work has appeared at LREC 2026, AACL-IJCNLP 2026, and workshops at ICML 2026 and EMNLP 2026, and he releases his datasets and code openly.
The question that motivates me
Affiliate Researcher at IUT Systems and Software Lab, focused on the languages, benchmarks, and model behaviors that mainstream LLM evaluations often skip.
No evaluated model exceeded 50% accuracy against 83.4% human performance.
Persuasive framing shifts coordinates by up to 0.57 units, and under-represented languages by 0.52.
Every model tested resisted persuasion better than it persuaded; even the weakest resister outranked the strongest persuader on Elo.
Across 439 games, containment is tightly clustered (σ = 1.9 Elo) while extraction varies widely (σ = 53.3 Elo); confirmation framing raises extraction odds 7.75×.
Output length alone explains 78.7% of between-system variance in human preference; matching outputs by length reverses the leaderboard.
The language a model is tested in shifts how strongly it is pulled toward the literal reading of an idiom.
Safety alignment in a low-resource language can hold in form while failing in meaning.
1,484 acts spanning 226 years, decomposed into 35,633 sections and 14,523 footnotes.
Recommends cosine similarity with CSLS as the default retrieval metric for multilingual embedding pipelines.
service · Bangladesh
May 2026 - Present
Academic member of Bangladesh's national AI olympiad. For the 2026 summer round, I helped set the questions for the national round.
research · Gazipur, Bangladesh
Sep 2026 - Present
Continuing research after my BSc on LLM evaluation and safety in low-resource languages, temporal reasoning in LLMs, and cross-lingual retrieval.
Aug 2025 - Sep 2026
Led my undergraduate thesis on the political behaviour of LLMs, which became LLM-Ideoplasticity (AACL-IJCNLP 2026); built the Bangla Bagdhara benchmark (LREC 2026); designed multi-agent evaluations of persuasion and containment (AREG at ICML, AIDG at EMNLP).
leadership · Gazipur, Bangladesh
Dec 2025 - Sep 2026
Led technical operations for the 12th National ICT Fest and built its platform, which handled 30,000+ registrations; convened and judged CodeSprint 2.0 and DevSprint 1.0; organised RISC-V Day.
Oct 2024 - Dec 2025
Technical Lead for CodeRush 2.0 (~1,500 registrations) and built its registration platform; set problems for the 11th National ICT Fest GameJam and Datathon.
leadership · Gazipur, Bangladesh
Dec 2025 - Sep 2026
Sep 2024 - Dec 2025
work · Dhaka, Bangladesh
Oct 2025 - Apr 2026
Designed database architecture and backend APIs for production ERP systems serving two retail brands, in Laravel and TypeScript, with CI/CD on Azure and GitHub Actions and end-to-end tests in Playwright.
BSc in Software Engineering