DEV Community: Harini

🌱 OutsideFix: What If AI Helped Us Step Away From the Screen?

Harini — Sun, 11 Oct 2026 13:45:45 +0000

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

What I Built

OutsideFix — One Photo. One Mission. One Small Action. 🌱

What if AI didn't just help us do more on our screens, but helped us step away from them?

That's the idea behind OutsideFix, an AI-powered outdoor action companion I built for the Touch Grass challenge.

Imagine noticing litter in a park but not knowing what practical action to take. OutsideFix lets you upload a photo of an outdoor issue and turns that observation into a manageable mission with actionable steps, an estimated duration, and safety guidance.

Here's how the experience works:

📸 Observe: Upload a photo of an outdoor issue.
🧠 Understand: Gemma helps analyze the image.
🎯 Take the first step: Get a practical, AI-generated outdoor mission.
🌿 Make it happen: Save the mission, take action when safe, and report your progress.

The goal is simple: reduce the distance between noticing a problem and doing something about it.

OutsideFix isn't about spending more time in another app. It's about making technology useful enough that you can close it and go outside.

Demo

🌐 Live application: https://outside-fix.vercel.app/
⚙️ Backend health check: https://outsidefix.onrender.com/api/health

The application is deployed with Vercel for the frontend and Render for the backend.

The best way to experience OutsideFix is to upload an image, generate a mission, and explore the actionable steps it provides.

Code

💻 GitHub repository: https://github.com/Harini-7228/OutsideFix

The project includes the frontend, backend API, AI integration, and mission-management functionality.

I'd love for the community to explore the code, try the application, and share suggestions for making it more useful.

How I Built It

I built OutsideFix as a full-stack application using:

React, Vite, and TypeScript for the frontend.
Node.js, Express, and TypeScript for the backend.
Gemma for image understanding and mission generation through a configurable model API.
MongoDB Atlas and Mongoose for mission persistence.
Vercel and Render for deployment.

The workflow starts when a user uploads an image. The backend validates the request, sends the image and prompt to the configured AI model, and uses the generated result to provide a structured outdoor mission. The application supports saving and managing missions.

I also incorporated input validation, upload limits, rate limiting, and safety-oriented logic.

One of the most important design decisions was focusing on the action that follows AI analysis. Recognizing an issue is useful, but helping someone identify a reasonable next step makes the experience more practical.

Why Does Open Innovation Matter?

Open innovation makes it possible for developers to experiment with AI beyond closed, general-purpose interfaces and build experiences around specific needs.

With Gemma, I could explore how an open-weight model can contribute to a focused application that connects image understanding with real-world action.

For OutsideFix, the interesting question wasn't simply, "Can AI describe this picture?"

It was, "Can AI help someone decide what to do next?"

Open models give developers opportunities to explore model choices, adapt workflows, and build specialized experiences around their own products. They also make it easier to learn by building, testing, and improving real applications.

OutsideFix is my exploration of that idea: using open AI to encourage practical action beyond the screen.

My Agent Session

OutsideFix uses a configured AI model API for image analysis and mission generation. I haven't included a DevRelay agent session because I don't have a saved session link to share.

Prize Categories

Gemma: Open-source AI integration for image understanding and outdoor mission generation.
MongoDB Atlas: Mission persistence and data management.
Render: Backend deployment.

🌎 One Photo. One Mission. One Small Action.

We often measure technology by how much it automates or how long people use it.

OutsideFix explores a different measure: does technology help someone do something meaningful in the real world?

A small action might not solve every problem, but it can make the first step easier.

I'd love to hear your thoughts:

If AI could help you take one small action to improve the world around you, what would you choose?

devchallenge #hf26challenge

I Built My Friend an AI Interviewer That Remembers Where He Struggles

Harini — Sun, 04 Oct 2026 22:32:06 +0000

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

🧠 What I Built

Most interview platforms give you questions and a score.

InterviewMirror remembers where you struggle.

I built InterviewMirror for my friend Tharun, who was preparing for technical interviews.

The problem was simple: practicing more questions doesn't necessarily help when the same concepts keep causing difficulty.

InterviewMirror is an adaptive AI interviewer that conducts technical interviews, evaluates answers, identifies knowledge gaps, remembers previous weaknesses, and uses that information to make future practice more personalized.

The goal is simple:

Make every interview more useful than the previous one.

🎤 Demo

Live Demo: https://interviewmirror-xlks.onrender.com/

Try the complete interview experience and see how InterviewMirror evaluates responses and provides personalized feedback.

The focus isn't just on getting a score.

It's on understanding what you need to improve before your next interview.

💻 Code

GitHub Repository: https://github.com/Harini-7228/interviewmirror

The complete project is publicly available for exploration.

🛠️ How I Built It

InterviewMirror combines open-weight AI, persistent memory, voice interaction, and cloud deployment to create a more realistic interview experience.

🤖 Gemma — The Intelligence Layer

Gemma powers the core AI experience.

It is used for generating interview questions, evaluating candidate responses, identifying weaknesses, and creating relevant follow-up questions.

The important part is that Gemma isn't simply used as a chatbot.

Its analysis directly contributes to the personalized interview experience.

🧠 MongoDB Atlas — The Memory Layer

InterviewMirror stores relevant interview information, including:

Interview history
Candidate responses
Strengths
Weaknesses
Performance information

This allows the application to retain useful context instead of treating every interview as a completely new session.

🔊 ElevenLabs — The Voice Layer

Technical interviews are conversations, not just text forms.

ElevenLabs gives the AI interviewer a voice, making the interaction feel more natural and closer to an actual interview.

☁️ Render — The Deployment Layer

InterviewMirror is deployed on Render and is publicly accessible through the live demo.

🌍 Why Does Open Innovation Matter?

For me, open innovation isn't simply about adding an AI model to an application.

It's about being able to build an experience around an open-weight model and experiment with how that model can be used in a real product.

With Gemma, InterviewMirror can use AI for more than generating answers.

It can analyze a candidate's performance, identify areas that need attention, and contribute to a more personalized practice experience.

This makes the AI a meaningful part of the product rather than a decorative feature.

🏆 Prize Categories

Gemma

Used for AI-powered interview generation, response evaluation, weakness detection, and personalized follow-up questions.

MongoDB Atlas

Used to persist interview history and candidate performance information.

ElevenLabs

Used to provide voice interaction for the AI interviewer.

Render

Used to deploy the publicly accessible application.

💡 What I Learned

Building InterviewMirror changed how I think about AI applications.

I initially thought:

"How can I build an AI interviewer?"

The more important question became:

"How can I build an interviewer that helps someone improve?"

That shift changed the entire project.

Instead of focusing only on generating questions, I focused on understanding the candidate, remembering useful context, and making practice more meaningful.

Building it for a friend also kept the project grounded in a real problem rather than just a technology demo.

❤️ Final Thought

An interview shouldn't end with a score.

It should leave you knowing:

What did I do well?

Where did I struggle?

What should I improve before my next interview?

That's what I built InterviewMirror to help answer.

Your previous interview changes your next interview. 🎯

Please Like, Comment and Share your thoughts!!