RESEARCH PROGRAMMER · DATA SCIENCE UNDERGRADUATE
Ahsan Habib Shihab
Al-Farabi Kazakh National University
Almaty, Kazakhstan
Research interests
LLM evaluation & benchmarking · Multilingual & low-resource NLP · Agentic systems & governance · AI reliability in sensitive domains
Education & research experience
B.Sc. in Data Science
Al-Farabi Kazakh National University · June 2027 (expected)
Research Programmer · Halyk Academy
Al-Farabi Kazakh National University
Model evaluation and experimental analysis under the national Kazakh-language LLM initiative (BR24993001). Research includes source-grounded historical knowledge evaluation and applied LLM reliability.
Published work
A. H. Shihab, A. Ospan, A. Mussa, and M. Mansurova. Same Problem, Different Answer: Accuracy Hides Cross-Lingual Inconsistency in Multilingual Inference. Proceedings of the 3rd Workshop on Language Understanding in the Human-Machine Era (LUHME 2026) · Association for Computational Linguistics · Budapest, Hungary · pp. 246–255 (2026).
Publisher recordA. H. Shihab, A. Mussa, A. Ospan, and A. Delavar. Circularity-Controlled Evaluation: An Independent Assessment Methodology for AI Governance and Oversight Mechanisms. 2026 IEEE 24th International Symposium on Intelligent Systems and Informatics (SISY) · Pula, Croatia · pp. 397–404 (2026).
10.1109/SISY71351.2026.11738459A. H. Shihab, A. Mussa, and M. Mansurova. How Many Advisors Are Enough? Predicting Useful Majority-Vote Ensemble Size under Correlated Decision Errors. 2026 IEEE 13th International Conference on Intelligent Systems (IS) · Varna, Bulgaria · pp. 1–7 (2026).
10.1109/IS71416.2026.11727194Aman Mussa, Zhanseit Tuimebayev, Madina Mansurova, Ahsan Habib Shihab. When Does Fine-Tuning Matter for a Low-Resource Language? Evaluating Open-Weight LLMs on Kazakh Benchmarks. Information · 17(9), 921 (2026).
10.3390/info17090921Ahsan Habib Shihab, Aman Mussa, Madina Mansurova. When Identical Conditions Yield Different Authority: Measuring Path-Dependence in Trust-to-Authorization Policies. MIWAI 2026 · Springer, LNCS · pp. 259–272 (2026).
10.1007/978-981-92-6510-7_20R. A. Druba, A. H. Shihab, A. Mussa, M. Mansurova. Privacy-Preserving Federated Learning for Post-Flood Security and Resource Planning in Bangladesh: A Review and Conceptual Framework. IEEE QPAIN 2026 (2026).
10.1109/QPAIN69676.2026.11546266A. H. Shihab, A. Mussa, and M. Mansurova. Towards Trustworthy AI in Energy Systems: A Risk-Aware Evaluation Framework for Safe Deployment in Critical Infrastructure. Engineering Proceedings, 160(1), 50 (2026).
10.3390/engproc2026160050Accepted work & presentations
How Many Prompts Certify a Ranking? A Generalizability Theory Protocol for Compute-Optimal, Dependable LLM Benchmark Evaluation. ICCBD 2026 (2026).
Presented. Proceedings and DOI are not yet confirmed.
Specification–Behavior Gaps in Trust-Based Authorization: A Gated-Quantity Inertness Diagnostic and Its Limits. CSAI 2026 (2026).
Accepted/camera-ready. Presentation mode and DOI are not yet confirmed.
Risk-Aware Evaluation of Domain-Adaptive Language Models in Sensitive Public-Sector Applications. AICT 2026 · 14–16 October (2026).
Presented. Publication status and DOI are not yet confirmed.
The Verdict Is in the Protocol: A Circularity-Controlled Evaluation Methodology for Runtime Governance of Agentic AI. IEEE Access ().
Accepted at IEEE Access. Publication details and DOI are not yet confirmed.
Selected projects
Sana
Core AI system lead · research methodology · evaluation design · technical coordination · Competition project · October 2026
Making official statistical microdata accessible through natural-language questions.
Al-Farabi AI Avatar
Evaluation corpus development · quantitative experiments · iterative refinement · Ongoing research
Evaluating historical knowledge with primary texts and interdisciplinary expert review.
Orbitum
Co-initiator & AI systems lead · Continuing project
From an EdTech hackathon idea to explainable competency and opportunity intelligence.
Windline
Hackathon participant & prototype contributor · Hackathon prototype · September 2026
Wind-power forecasts that account for what weather information was actually available at the time.
CivicShield AI
Independent research · Research framework in development
A reliability-aware evaluation framework for behavior that headline accuracy can miss.
Teaching, service & speaking
AI Summer School Instructor
Al-Farabi Kazakh National University
Delivered hands-on multi-agent AI workflow sessions using Pronto Sage to visiting students from China, including participants from Shanghai Jiao Tong University.
Conference peer reviewer
IEEE AICT 2026 · IEEE SISY 2026
Reviewed five papers for AICT 2026 and completed peer-review service for SISY 2026.
Invited speaker & institutional representative
II International Congress of Young Scientists of Kazakhstan
Nominated by the department to represent KazNU. Contributed to a panel on interaction between science, business, and society, discussing domain adaptation and localized language models.
Climate Change AI Summer School participant
Climate Change AI · virtual
Explored connections between AI methods, climate resilience, and privacy-aware flood-related planning.
Recognition
3rd place, AI for Official Statistics track, STAT.DATATHON-2026 — Team Sana, October 2026.
Best Position Paper Award, Silk Way Model United Nations, 2024.
Tools & methods
Evaluation & analysis: Experimental design, model benchmarking, failure-mode analysis, repeated-query testing, reproducible research.
Language models: Hugging Face Transformers, vLLM, LM Evaluation Harness, QLoRA, multilingual NLP, retrieval-augmented generation.
Data & machine learning: Python, SQL, R, NumPy, pandas, SciPy, scikit-learn, PyTorch.
Research workflow: Linux, Git, Jupyter, tmux, GPU workflows, experiment logging, LaTeX.
Published work, accepted papers, presentations, and work under review are listed separately. Updated 3 November 2026.