README.md

---
title: Echo Ledger
emoji: 🎙
colorFrom: green
colorTo: green
sdk: gradio
sdk_version: 6.18.0
app_file: app.py
pinned: false
tags:
- hackathon
- build-small-hackathon
- backyard-ai
- best-use-of-modal
- best-demo
- best-agent
license: mit
---

# Echo Ledger

Talk. Get stuff done.

You say "email Sarthak about the logs, book a team sync for Thursday, draft an investor outline" -- Echo Ledger hears it, pulls out every task, and hands you ready-to-use drafts. No typing. No forgetting. No $0.02 per token.

[Demo video](#) · [Social post](#)

---

## What it does

Record a voice memo. Whisper transcribes it. Gemma 4 12B reads the transcript and lists every task you mentioned. You pick which ones to automate. It generates the draft -- email body, calendar invite, presentation outline -- and gives you a one-tap link to send or download it. Everything lands in an activity log you can scroll through later.

That is the whole thing.

---

## How to run it

```bash
git clone https://github.com/davezaxh/echo-ledger
cd echo-ledger
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env   # add your HF_TOKEN
python mcp_server.py   # terminal 1
python app.py          # terminal 2 -- open localhost:7860
```

Needs Apple Silicon for local inference. Swap `USE_MODAL=true` in `.env` to run on Modal instead.

---

## Models

| Model | Params | Role |
|---|---|---|
| Whisper medium | 769M | Transcription |
| Gemma 4 12B | 11.95B | Extraction + drafting |
| Total | ~12.8B | Under the 32B cap |

---

## Built with

Gradio 5 -- faster-whisper -- mlx-lm -- Modal A10G -- SQLite -- FastAPI

---

*Built for the Hugging Face Build Small Hackathon. Runs on a Mac Mini M4. No subscriptions harmed in the making of this app.*