Project Deep Dive — 2025

Guideon

An AI learning assistant that recommends personalized learning paths using a locally hosted LLM (Llama 3 via Ollama) with hybrid filtering aligned to the Philippine Skills Framework.

Role: Full-Stack Developer
Year: 2025
Stack: Django, React, PostgreSQL, pgvector, Llama 3, Ollama
Links: GitHub

Guideon is an intelligent learning assistant that recommends personalized learning paths using a locally hosted LLM — built as our BS Computer Science undergraduate thesis at MSU-IIT.

Situation

Filipino learners and jobseekers often don't know which skills to learn, in what order, to reach a target career. The Philippine Skills Framework (PSF) maps those skills, but it's hard for an individual to apply — the real difficulty is knowing which actual course satisfies which PSF requirement. On top of that, cloud LLMs raise privacy and cost concerns for a student-facing tool.

Task

For our group undergraduate thesis, we set out to build and formally evaluate a recommender that turns the PSF into concrete, personalized learning paths — with the AI running entirely on local infrastructure. I worked as a full-stack developer across the system.

Action

Integrated a locally hosted Llama 3 via Ollama — no user data sent to cloud LLMs
Built hybrid filtering that combines collaborative and content-based recommendations
Aligned recommendations to the Philippine Skills Framework so courses map to real PSF skills
Developed the full-stack architecture: Django backend, React frontend
Used PostgreSQL with pgvector for embedding storage and similarity search

Result

CUQ usability score of 71.8 (SD ±11.7) in formal evaluation — high user satisfaction
Working PSF-aligned recommendations from a fully local LLM pipeline
Completed and defended as our undergraduate thesis