Aakkash MuthukumarRésumé ↗

Case study · Fall 2025

STEM Forge

An award-winning, local-first learning platform that generates adaptive quizzes, coding problems, and custom courses.

Discipline
AI education · Full-stack product
Technology
Python · FastAPI · React · Gemma 3n · Ollama

STEM FORGE

What do you want to learn?

Adaptive quizCoding labCustom course
LOCAL MODEL · READY

Overview

Built for MESA U Hacks 2.0, STEM Forge pairs a FastAPI backend and React interface with Gemma 3n through Ollama so the full learning experience can run on-device.

Best AI IntegrationRecognition
Fully localExecution
3 adaptiveLearning modes
HackathonBuild format

System map

How the pieces connect

01Learner request
02FastAPI · Gemma · Ollama
03Quiz · lab · course

The challenge

What made the problem worth solving

AI learning tools often assume a persistent cloud connection and optimize for answer generation. We wanted a tool that could support active learning even where connectivity or privacy made cloud dependence a poor fit.

The approach

Turning the problem into an engineering plan

  1. 01

    Designed multiple learning modes for quizzes, coding practice, and on-demand course generation.

  2. 02

    Connected a React frontend to a FastAPI service running Gemma 3n locally through Ollama.

  3. 03

    Kept the complete inference path on-device so learners could use it without reliable internet.

Result

STEM Forge received the Best AI Integration Award and reinforced a theme that now appears throughout my work: useful AI should meet people where they are, not require ideal infrastructure.

Next case studyCodivus→