Jomari Manto

Final Project

Finals · IEEE Report

Bumi: a multilingual RAG assistant for university student guidelines

Bumi, the Student Handbook Oracle, is a retrieval-augmented assistant grounded in three institutional documents: the College Student Handbook, the College General Information Bulletin and the Student Virtual Learning Code of Conduct. It responds in English, Taglish or Filipino, cites the page from which each answer is drawn and, where the documents do not address a query, refers the user to the appropriate office rather than speculating. The accompanying report describes the system's design, implementation and evaluation.

Read the Report (PDF, opens in a new tab) Access the Assistant (opens in a new tab)

What Bumi Does

1. 
   01

   Multilingual Support

   Responses are generated in English, Taglish or Filipino, either by user selection or by detecting the language of the query.

2. 
   02

   Source-Cited Responses

   Each response references the specific document and page from which the information was retrieved.

3. 
   03

   Referral to University Offices

   Where the source documents do not address a query, Bumi states this explicitly and directs the user to the appropriate office.

4. 
   04

   User Feedback Mechanism

   Users evaluate each response for helpfulness and accuracy and may submit an anonymized transcript for review.

The Course in Three Terms

1. 01

   Prelim

   NLP and the AI hierarchy Artificial intelligence is a field of study rather than a tool, and its applications were embedded in daily life long before the emergence of generative AI.
2. 02

   Midterm

   LLMs, traditional ML and LLMOps Large language models remain a form of machine learning, and their reliability depends on the context they receive and the operational discipline surrounding them.
3. 03

   Finals

   Retrieval-Augmented Generation RAG is the practical answer to the lesson that an LLM is only as reliable as the context it receives.