My mom struggled to write daily sales speeches. So I built an AI agent for her.

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

What I Built

Every single day, my mother works at a sports-betting booth. Part of her job requires pitch-selling match odds to customers with an engaging speech in the morning and compiling a concise sales report at night.

She isn't very tech-savvy, so she used to rely on my siblings or me to manually search the web for match standings, team streaks, and statistics. I wanted to build a lightweight, zero-friction solution that gives her complete independence without making her navigate complex menus or write prompts.

I created TeApuesto Assistant: a mobile-first PWA where my mom simply uploads a picture of the day's odds board. The AI agent extracts the matches, fetches live sports statistics via SerpApi, and automatically generates:

1. Morning Pitch Speech: An engaging, easy-to-read sales pitch for her booth's customers.
2. Evening Daily Report: A structured summary of the daily betting performance.

It runs live today and she uses it every single day on her phone.

Demo

Watch the demonstration video:

See the PWA deployed: teapuesto.juanhuanaco.tech

Code

The source code is open-source and available on GitHub:
👉 github.com/juanhuanaco/teapuesto-agent

How I Built It

The app is built as a clean, minimal TypeScript full-stack solution designed for high performance and low operational cost:

• Mastra: The agent orchestration framework used to route tools, manage state, and handle structured schemas for agent responses.
• Gemma (Open-Weight Model): Powered by Google's Gemma model served via Featherless.ai API, providing strong reasoning and multimodal understanding without heavy local hardware requirements on her phone.
• SerpApi: Integrated directly as a Mastra tool to ground the LLM with fresh, real-time web results for sports standings and team statistics.
• Frontend: Lightweight Vite + Tailwind CSS PWA featuring Base64 image uploading, tab switching between endpoints, and one-tap clipboard copying.
• Backend: Express 5 on Node.js, deployed seamlessly to Render.

Why Does Open Innovation Matter?

Open innovation and open-weight models (like Gemma) combined with open orchestrators (like Mastra) make AI accessible for real, hyper-local problems.

Instead of relying on rigid proprietary SaaS tools or expensive subscription black boxes, open tools allowed me to craft a tailored, single-purpose micro-agent specifically for my mom's daily job. Open AI democratizes building hyper-practical tools for real people who aren't tech-savvy.

The Lazy Payoff

This project saves roughly 30 to 45 minutes every day. My mom no longer needs to wait for us to search stats or type summaries. She uploads a photo, taps one button, and gets her morning speech or nightly report in seconds. It turned a daily chore into a two-tap task.

Prize Categories

• Best Use of Gemma: Leverages Google's Gemma open-weight model served via Featherless.ai for multimodal analysis and speech generation.
• Best Use of Mastra: Uses Mastra (@mastra/core) as the core agent harness to orchestrate LLM execution and sports search tools.
• Best Use of SerpApi: Grounded in real-time sports data and news extracted via SerpApi queries.
• Best Use of Render: Hosted end-to-end on Render (Express 5 backend + static PWA bundle).
• Best Use of a .tech Domain: Deployed live on a custom subdomain at teapuesto.juanhuanaco.tech.