Research · source-grounded generation
Al-Farabi AI Avatar
Evaluating historical knowledge with primary texts and interdisciplinary expert review.
At Halyk Academy, I contribute to the evaluation and refinement of a retrieval-augmented generation system for knowledge about Al-Farabi and his intellectual works.
01 / THE QUESTION
How do we evaluate a generated historical answer against primary texts and the judgment of domain specialists?
Halyk Academy · Al-Farabi Kazakh National University.
- 01Primary texts
- 02Question–answer corpus
- 03Expert review & experiments
Reference answers are reviewed by specialists in history, philology, and theology.
02 / THE WORK
My contribution
Develop a domain-specific question-answer evaluation corpus grounded in primary texts.
Work with reference answers reviewed by specialists in history, philology, and theology.
Conduct quantitative experiments to identify weaknesses in retrieval and answer generation.
Translate evaluation findings and expert feedback into targeted system improvements.
Methods & focus
- Retrieval-augmented generation
- Expert-validated benchmarking
- Source-grounded evaluation
03 / EVIDENCE & LIMITS
What the work supports
Research and methodology papers are being developed to document the approach and findings.