HOMI - Helpful Online Mentor Interface

A study assistant for IT students built on a fine-tuned LLM. It answers questions, writes study material, generates practice tests, and tracks attendance and assignments.

Description

HOMI is a full-stack study assistant for BSc IT students, built on a fine-tuned large language model (LLM). It writes study material tailored to you, generates multiple-choice questions that match your level, and answers your doubts through a chatbot as they come up. It also keeps track of your attendance and assignments. We built it because regular classes don't adapt to how each student is doing, and we wanted studying to feel more personal and interactive.

Media

1 of 10

HOMI answering a question about the 8051's timer registers.

Features

Fine-Tuned LLM

A customized LLM based on DeepSeek does the AI work, like answering questions and generating tests.

Chatbot

You can chat with the model to ask questions or generate study material, and sort everything it gives you into folders.

Attendance and Assignment Tracking

You log your attendance for each subject, and HOMI works out how many more classes you can miss and still stay above your target. You can also log assignments and use the site as a checklist for them.

Test Generation

You can test yourself on a topic you've studied with graded multiple-choice tests or with flashcards.

Secure Authentication

Logging in uses secure authentication, and a PostgreSQL database stores user details, questions, tests and the rest.

Key Responsibilities

  • Worked as project manager for the team of 6 people building it.
  • Trained the LLM (based on DeepSeek R1) on cloud and local machines.
  • Collected the training data from textbooks, using manual scanning, distillation and several scraping techniques.
  • Built several backend features, mostly the parts that talk to the model.
  • Helped deploy the backend to Hugging Face Spaces, and uploaded the model to Hugging Face.
  • Helped design several UI components and pages.
  • Paid for development myself until our college gave us a grant.

Additional Information

  • This was my undergrad capstone. It took seven months from the first idea to the final presentation, which happened before I graduated.
  • We also planned a mental health guide, a study roadmap generator and PDF export for chats, but ran out of time.
  • The project got a grant from outside investors, who plan to keep building on it.
  • Our research for it turned into 3 papers. One was presented at an IEEE conference, one was published in IJERT, and one is still under review.
  • During the final assessment, an examiner from IIT Bombay (Indian Institute of Technology Bombay) singled out the project for its technical depth and originality.