Raha, Torabihaghighi (2026) A Proof-of-Concept RAG Recommendation System for Scientific Equipment. Bachelor's Thesis, Artificial Intelligence.
|
Text
BachelorsProjects5618762.pdf Download (646kB) | Preview |
Abstract
This paper explores Retrieval-Augmented Generation (RAG) as a method for recommending scientific equipment from a structured database based on natural-language problem descriptions under limited expert-labeled data. The study uses an equipment database of 8,601 records and evaluates the system primarily on two synthetic query-answer datasets and a manually written (and later expert-reviewed) evaluation set. The system combined retrieval, LLM-based reranking, prompt engineering, and output validation. The results show that such a system can retrieve and rank equipment candidates under the experimental setup and can enforce database-grounded outputs. The retrieval experiments suggest that the dense retriever is more effective than the lexical retriever for the present task. The fine-tuned retrieval results imply that fine-tuning the dense retriever, either on its own or as part of a hybrid, may improve retrieval performance. However, the results should be interpreted with caution due to reliance on synthetic data and limited evidence of generalization to real user queries.
| Item Type: | Thesis (Bachelor's Thesis) |
|---|---|
| Supervisor name: | Dibangoye, J.S. |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Bachelor's Thesis |
| Language: | English |
| Date Deposited: | 09 Jul 2026 12:43 |
| Last Modified: | 09 Jul 2026 12:43 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/37408 |
Actions (login required)
 |
View Item |