Ayuda: Your Time, for the Greater Good!

Abhimanyu Mathur — Sun, 11 Oct 2026 15:01:02 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

Transform your screentime from doomscrolling, and explore greater good tasks near you. I bring you, Ayuda: voice-first, location-aware volunteering app.
You say what you can offer ("two hours on Sunday morning, I can teach English to kids, no money"), and Ayuda finds verified organisations nearby that need exactly that. Then it gets out of your way so you can go and help.

Demo

Ayuda Video Demo.mov - Google Drive

drive.google.com

Code

supermanyu90 / ayuda

Voice-first, location-aware volunteering: open-weight Gemma 3 4B + ElevenLabs + PostGIS

Ayuda

Give your time. Find where it matters.

Ayuda is a voice-first, location-aware volunteering app. You say what you can offer ("two hours on Sunday morning, I can teach English to kids, no money"), and Ayuda finds verified organisations nearby that need exactly that. Then it gets out of your way so you can go and help.

Built for the DEV Hacktoberfest Open-Source AI Challenge, Week 1: Touch Grass: open-weight AI that gets people off the screen and into the world.

Most AI products optimise for more screen time. Ayuda optimises for the opposite: the shortest safe path from one sentence to real-world action Success metric: real-world volunteering minutes per minute spent in Ayuda. The app measures it: visible-tab time is counted on-device from the start of a plan to the commitment, and the exit screen tells you, e.g. "You spent 1 min 52 s on your screen to…

View on GitHub

How I Built It

Ayuda uses Gemma 3 4B, Google's open AI model, which runs on a local computer through Ollama. It helps Ayuda understand and respond to users in three main ways:

Understands what you say: No surprises here, pretty vanilla with AI these days.
Helps NGOs create listings It prepares draft listings based on information provided by an NGO. An administrator reviews and approves each listing before it becomes visible to users.
Creates short voice responses: It writes a simple, two-sentence summary that Ayuda can read aloud to the user.

To keep responses consistent and reliable, Ayuda uses additional checks to ensure the AI follows a predefined format and that its responses meet strict validation rules.

Note that, Ayuda does not let AI decide which organisations are real or trustworthy. The AI only helps understand what users say. Information about organisations, their locations and their verification status comes from a structured database. Google Reviews can be added here.

Why Does Open Innovation Matter?

Open weights were what made Ayuda work. Four things it could do that a closed API wouldn't have allowed:

The request stays on your machine. The one exception is ElevenLabs voice, and typing or tapping keeps the whole AI path local.
I could find the failures, fix them, and prove the fix.
Communities can adapt it themselves. With a closed API, they could only ask the vendor for that feature.
A small NGO can afford it. A 4B model runs on a laptop. Volunteering tools mostly serve organisations with no budget, and they shouldn't depend on per-toke on an API key that could be revoked.

Prize Categories

(a) Gemma
(b) ElevenLabs

OutingFit Mumbai: Dress for the whole outing, not just the forecast.

Abhimanyu Mathur — Sun, 04 Oct 2026 15:53:46 +0000

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

What I Built

OutingFit Mumbai helps you decide what to wear and what to carry for an outing in Mumbai, based on the actual forecast for when you'll be out. It works by voice or text in English, Hindi or Marathi.
I have built this application for my lady friends who get confused that should they wear dress, jacket, carry a raincoat or sunglasses.

Demo

OutingFit

Code

supermanyu90 / OutingFit

Tell OutingFit where you’re going—in your language—and get an outfit and packing checklist grounded in the weather for your outing

OutingFit

Weather-grounded outfit and packing advice for outings in Mumbai, in English, Hindi and Marathi, by voice or text.

 speak / dictate / type ──► ElevenLabs (agent · Scribe STT) ──► outing form (visible, editable)
                                                                   │  confirm place + date/times
                                                                   ▼
             Open-Meteo forecast for the outing window ──► decision rules (config.ts) ──► Gemma
                                                                   │                     (explains
                                                                   ▼                      localises)
              Wear / Carry / Check cards + spoken summary ◄── validation (schema, coverage, numbers, script)
                                                                   │
                                          ElevenLabs TTS / agent speaks exactly that summary

How it works

Input. The user talks to an ElevenLabs agent, dictates (Scribe v2 speech-to-text, with an editable transcript) or types. Gemma extracts the fields from dictated or typed text. The agent fills them itself through client tools. Either way the form on screen shows exactly what was understood.
Confirmation. Weather is not fetched until the destination is unambiguous and the date and times are confirmed. "Bastian" matches…

View on GitHub

How I Built It

1.Input: An ElevenLabs agent, Scribe speech-to-text (you can edit the transcript) or typed text. Gemma pulls the venue, date, times and preferences out of free-form Hindi, Marathi or English. Everything it understood shows up in an editable form before anything else happens.
2. Weather: The Open-Meteo hourly forecast for the outing window.

*For Logic and Control *, a deterministic rules engine with configurable thresholds turns the weather and what you told it into prioritised items, each with its evidence.

3. Gemma 4 (gemma-4-26b-a4b-it / 31b-it): writes the titles, explanations, outfit idea and a 60-word spoken summary. It can't add, drop or re-rank items.
4. Voice out: ElevenLabs again.
Gemma runs on the Gemini API in production, with a local Gemma 3 4B fallback through Ollama. Stack: TypeScript, Express, React 19, Vite, Tailwind, Zod. It's deployed on Render. 31 unit tests and 9 end-to-end scenarios cover the decision rules, the validators, the time-zone handling (including returns after midnight) and API-failure states. All of them pass.

Why Does Open Innovation Matter?

Apart from being Global Financial Hub, Mumbai rules with classy restaurants and has a globally recognized social klout. It's going to rain, or will I sweat my brains out: addressing the weather issue is always a concern in this happening city. One cannot browse multiple sources, predict the weather, and choose clothing. Here comes OutingFit. It takes on the recent extreme weather changes in the coastal city of Mumbai. A voice-enabled assistant that checks hourly weather for your outing and provides evidence-backed Wear / Carry / Check cards, a spoken summary, and outfit ideas.
You can choose the best from what you have, and no location-based suggestions from Amazon or Flipkart bother you to buy more and more.

Prize Categories

Featured categories

Best Use of Render
Best Use of Gemma Partner categories
Best Use of ElevenLabs
Best Use of Sentry Agent Tracing
Best Use of SerpApi

DEV Community: Abhimanyu Mathur

Ayuda: Your Time, for the Greater Good!

What I Built

Demo

Ayuda Video Demo.mov - Google Drive

Code

supermanyu90 / ayuda

Voice-first, location-aware volunteering: open-weight Gemma 3 4B + ElevenLabs + PostGIS

Ayuda

How I Built It

Why Does Open Innovation Matter?

Prize Categories

OutingFit Mumbai: Dress for the whole outing, not just the forecast.

What I Built

Demo

Code

supermanyu90 / OutingFit

Tell OutingFit where you’re going—in your language—and get an outfit and packing checklist grounded in the weather for your outing

OutingFit

How it works

How I Built It

Why Does Open Innovation Matter?

Prize Categories