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

What I Built

I built EchoNote AI for my friend Anshumita Ray.

Anshumita records her college professor's lecture on her phone during class. She does this because she cannot catch everything in real time. Sometimes a side conversation distracts her. Sometimes she forgets an important word, or misses a whole part of the lesson. Later, she has to listen to hours of audio again just to write her notes.

EchoNote AI fixes this. She uploads her recording or records live in the browser. The app writes the full transcript, then makes a clean summary with key definitions and exam questions. If she forgot something, she can ask the AI a question about the lecture and get an answer based only on what the professor said.

It runs fully on a normal laptop. No API keys. No internet after setup. Her notes are stored in an encrypted local database on her own device.

Demo

Code

How I Built It

Every important part is open source:

1. faster-whisper turns speech into text. It supports 99 languages and handles mixed language lectures, which is common in our classrooms.
2. Gemma 3 1B (GGUF, Q4) writes summaries, exam questions, and answers chat questions. It runs through llama-cpp-python and needs only about 800 MB of RAM.
3. FastAPI powers the backend.
4. SQLite with Fernet encryption keeps transcripts, summaries, and chats private.
5. HTML, CSS, and JavaScript make a simple three column studio screen.

The app checks the hardware when it starts. It uses an NVIDIA GPU or Apple Metal if they exist, and falls back to CPU on any laptop with 4 GB RAM. I picked the small 1B model on purpose, because a student laptop must be able to run it.

It also detects the type of audio. A college lecture gets exam questions. A meeting gets action items. A recipe gets ingredients and steps.

Why Does Open Innovation Matter?

Here is the thing. A closed API would have failed Anshumita in three ways.

1. Cost. A student cannot pay for every hour of lecture audio.
2. Privacy. Her professor's lectures and her own voice notes should stay on her device, not on someone else's server.
3. Access. The app works offline, so bad internet does not stop her from studying.

Open weight models made all three possible. I chose a model small enough for her laptop, ran it myself, and changed how it writes notes for each type of audio. With a closed API I could not have done that for free.

Prize Categories

Best Use of Gemma. EchoNote AI runs Gemma 3 1B locally on CPU. It writes all summaries and exam questions, and it powers the chatbot that answers from the lecture transcript.