TOEFLGemma AI — An Offline Open-Source TOEFL iBT Mentor for My Colleague

DinitaKs — Sun, 04 Oct 2026 09:54:03 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

I built TOEFLGemma AI—an offline, privacy-focused TOEFL iBT study mentor powered by Google's Gemma model.

I designed this tool for a close colleague and friend who is currently preparing for the TOEFL iBT exam. Studying for TOEFL iBT requires practicing complex academic English, structured templates for Speaking and Writing tasks (including the 'Writing for an Academic Discussion' format), and specific note-taking strategies.

However, commercial AI tutors often require constant internet connectivity and expensive monthly subscriptions. Since my friend frequently works in locations with spotty Wi-Fi connections and is on a strict study budget, I built a 100% local solution that turns a standard laptop into a dedicated, offline TOEFL mentor with zero subscription fees and complete data privacy.

Demo

GitHub Repository: my Repo :D

Code

The repository contains the custom Modelfile configuration that shapes Gemma 2 into an authoritative yet encouraging TOEFL iBT tutor, along with setup guides and local UI integration steps.

You can view the full repository here:

atDnK / edugemma-ai

Offline educational assistant powered by Google's Gemma model

📚 TOEFLGemma AI — Offline TOEFL iBT Study Mentor

TOEFLGemma AI is a lightweight, privacy-first, offline study assistant powered by Google's Gemma 2 (2B) open-weight model. It is built specifically to help students and self-learners master TOEFL iBT skills without relying on paid subscriptions or active internet connections.

✨ Features

100% Offline Inference: Operates locally via Ollama with zero internet dependency.
Task-Specific Templates: Generates structured frameworks for Speaking (Tasks 1–4) and Writing (Integrated & Academic Discussion).
Academic Vocabulary Tutor: Breaks down complex reading passages into simplified terms and context examples.
Zero Operating Cost & Full Privacy: All practice essays, notes, and queries remain 100% private on your machine.

How to Run Locally

Install Ollama.
Pull the Gemma 2 model in your terminal ollama run gemma2:2b
Load the custom system prompt from Modelfile into your preferred local Web UI or create a custom Ollama model ollama create toeflgemma…

View on GitHub

Here is the exact Modelfile running locally on top of Gemma:

FROM gemma2:2b

PARAMETER temperature 0.3

SYSTEM """
You are TOEFLGemma AI, a specialized offline TOEFL iBT study mentor and tutor.
Your job is to help users master TOEFL iBT skills: Reading, Listening, Speaking, and Writing.

Rules:
1. Provide structured templates for Speaking and Writing tasks (including 'Writing for an Academic Discussion').
2. Explain complex academic vocabulary in simple terms with context examples.
3. Offer actionable note-taking strategies for Listening passages.
4. Keep feedback encouraging, precise, and focused on TOEFL iBT rubrics.
"""

How I Built It

I built TOEFLGemma AI using a completely open-source, local stack:

Open-Weight Model: Google's Gemma 2 (2B) served locally via Ollama, allowing smooth, low-latency inference on regular consumer laptops without demanding high-end GPUs.
System Prompt Tuning: Custom system prompts configured with low temperature (0.3) to prioritize factual accuracy and structured learning outputs based on official TOEFL iBT rubrics.
Local Frontend Interface: Integrated with AnythingLLM / Page Assist to provide a clean, distraction-free chat interface for full study sessions.

Why Does Open Innovation Matter?

Open innovation and open-weight AI models like Google's Gemma make local solutions genuinely accessible:

Zero Internet Reliance: My friend can practice speaking and writing responses anytime, anywhere—even during travel or network outages.
100% Free & Barrier-Free: Eliminates costly API keys and SaaS monthly fees for students preparing for expensive standardized tests.
Complete Data Privacy: Practice essays, personal notes, and draft responses never leave the local device, ensuring total privacy.

Prize Categories

Overall / Build for a Friend (My Beloved Sis, Mot<3)

DEV Community: DinitaKs