DEV Community: Somesh Kumar Singh

InstaPilot AI

Somesh Kumar Singh — Fri, 16 Oct 2026 20:11:08 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

I built InstaPilot AI, a full-stack content management platform featuring an AI-driven "human-in-the-loop" approval workflow for Instagram.

Who would pay for it? Social media managers, indie hackers, and digital marketing agencies.
What problem does it solve? Scaling social media content creation usually means either spending hours brainstorming captions and designs, or risking brand alignment by letting fully automated AI tools post directly to channels without review. InstaPilot AI solves this by automating the heavy lifting (image and caption generation) while enforcing a strict human-in-the-loop gate (Draft ➡️ Approved ➡️ Published) so creators maintain total editorial control before anything goes live.

Code

You can check out the complete open-source repository on GitHub:
👉 GitHub - instapilot-ai

How I Built It

InstaPilot AI is built as a modern full-stack application using open technologies:

Backend: Python 3.13, FastAPI, Pydantic v2, and Uvicorn for a high-performance asynchronous REST API.
Frontend: React, Vite, and Tailwind CSS styled with a modern glassmorphism dark-mode UI.
AI Engine & Storage: Integrated with Pollinations AI for dynamic, prompt-driven visual and text asset creation, backed by a lightweight local JSON persistence store (posts_db.json) and static asset mounting.
Workflow Architecture: Implements a proxy bridge (Vite to FastAPI) ensuring seamless cross-origin communication, live backend health telemetry, and stateful human approval gates.

Why Does Open Innovation Matter?

Open innovation and open-source tooling were foundational to building InstaPilot AI. Relying on closed, rigid ecosystems would have locked down how content states are managed, stored, and filtered. By leveraging FastAPI, React, and open AI media endpoints, I was able to customize the exact workflow logic, maintain local ownership of campaign data via transparent file storage, and bypass restrictive vendor constraints. Open innovation made it possible to build a production-ready side-hustle tool entirely from scratch with zero licensing barriers.

My Agent Session

Built with the assistance of AI coding agent workflows to rapidly scaffold the FastAPI backend and Tailwind React dashboard.

My First User

Tested locally with a fellow content creator who loved the glassmorphism UI and the peace of mind offered by the explicit approval gate before hitting publish!

🌿 IRL Clash — Go Outside. Take a Shot. Win the Battle.

Somesh Kumar Singh — Thu, 08 Oct 2026 19:02:14 +0000

IRL Clash — Go Outside. Take a Shot. Win the Battle.

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

What I Built

I built IRL Clash, a real-world multiplayer photography battle designed to get people away from their screens and into the outdoors.

Instead of joining a traditional game lobby, players simply enter a nickname and share their location. When two active players are close enough, the backend automatically matches them into a battle.

Each battle has 5 rounds, with 2 minutes per round.

For every round, players receive a random outdoor photography mission such as:

🍃 Find an interesting leaf
🌳 Capture an unusual tree
🚲 Find a bicycle
🌸 Find a flower
💧 Capture a water scene
🪨 Find an interesting rock
🌑 Take an interesting shadow shot
🪑 Find an outdoor bench
☁️ Capture an interesting cloud
🏛️ Photograph an architectural detail

Players can take a photo directly with their phone camera or select one from their gallery.

After the fifth round, the submitted images are evaluated by the AI referee. Each image receives a score out of 100 along with a short description and explanation. The five round scores are combined to determine the winner.

The goal is simple:

The screen isn't the game. The real world is.

IRL Clash is designed for friends, college students, meetups, and anyone who wants a more playful reason to go outside.

Demo

🌐 Live Backend: https://irl-clash.onrender.com/

🎮 Live Frontend: https://irl-clash.vercel.app/

The game works best on a mobile device because players can use GPS and the camera directly from their browser.

Code

💻 GitHub: https://github.com/Somesh-coding/IRL-Clash
The project is built as a full-stack application:

IRL Clash
│
├── Frontend
│ ├── React
│ ├── Vite
│ └── Browser Camera + Geolocation APIs
│
└── Backend
├── Java 21
├── Spring Boot
├── REST APIs
├── In-memory multiplayer state
└── AI image judging

There is intentionally no database for this version. Active players, battles, photos and scores are kept in memory, making the project lightweight and easy to run for a demo.

How I Built It

The frontend is built with React + Vite and provides the complete game experience, including:

Player registration
Browser geolocation
Automatic nearby-player matching
Five-round battle UI
Countdown timers
Camera capture
Gallery uploads
Live opponent status
Final score screen

The backend is built with Java 21 and Spring Boot.

It handles:

Player registration
Location updates
Distance calculation
Automatic player matching
Battle creation
Server-side round timing
Image uploads
AI evaluation
Score calculation
Winner determination

The application uses a configurable matching radius and a 120-second server-side timer for every round.

The image judging system sends the submitted images together with their corresponding missions to the AI referee. The referee returns structured scores, descriptions and reasons, which the backend uses to calculate the final result.

The backend is deployed on Render, while the frontend is deployed separately.

Why Does Open Innovation Matter?

For this project, the important idea is that the AI referee should not be the part that keeps people staring at the screen.

The AI is there to evaluate what players actually find outside.

Open AI tooling makes this kind of experiment easier to build, modify and extend. The architecture keeps the AI judging layer separate from the game itself, so the judging model can be replaced without redesigning the multiplayer system.

That opens the door to future versions using open-weight multimodal models and local inference.

For example, a future version could allow the AI referee to run locally on a laptop, edge device or phone. That would make it possible to keep players' outdoor photos and location-related data under their own control, while also allowing developers to experiment with different models and judging criteria.

The broader idea is:

Use AI to reduce the time people spend interacting with the application, not to increase it.

The app gives players a mission, gets them outside, and lets the real-world activity become the main experience.

My Agent Session

I used AI-assisted development during the implementation of IRL Clash for architecture, debugging, frontend/backend development and deployment troubleshooting.

Prize Categories
Overall — Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

Partner category: Best Use of Render

IRL Clash uses Render to deploy the Spring Boot backend and expose the multiplayer game APIs.

What's Next?

Some ideas for future versions:

Open-weight multimodal AI running locally
Persistent leaderboards
Team battles
More mission types
Friend challenges
Anti-cheat image verification
Better location privacy
Offline/edge AI judging
Public outdoor challenges and events

For now, the rule is simple:

Find someone nearby. Go outside. Take the shot. Win the clash. 🌿📸🏆

Study Buddy: a retro AI study companion built for a friend 🎓

Somesh Kumar Singh — Mon, 05 Oct 2026 01:25:00 +0000

devchallenge #weekendchallenge #hf26challenge #opensource

Hacktoberfest Weekend Challenge: Build for a Friend Submission 🤝
WebPage Link: https://study-buddy-u5vl-git-main-someshsingh2206-9453s-projects.vercel.app/

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

What I Built

Study Buddy is a retro-style AI study companion designed to make studying a little less overwhelming. I wanted to build something useful for a friend who has to go through lengthy notes, prepare for exams, and figure out what to revise next.

Instead of switching between different tools for reading notes, making quizzes, and planning study sessions, Study Buddy brings these tasks together in one place.

You can upload study material, ask questions about your notes, generate practice quizzes, create a study plan, and keep track of your progress. The goal is to spend less time organizing study material and more time actually learning it.

I built the interface with a nostalgic retro aesthetic because studying doesn't have to feel like staring at another boring productivity dashboard.

I also wanted the project to be accessible and easy to try, without requiring users to create an account or set up a database just to get started.

Code

View Study Buddy on GitHub:https://github.com/Somesh-coding/study-buddy

Replace the link above with your actual public repository before publishing.

How I Built It

Study Buddy combines a Java backend with a modern React frontend.

Frontend: React and Vite, with a retro-inspired interface designed to make studying feel approachable.
Backend: Java and Spring Boot, providing REST APIs for the study features.
AI: Hugging Face Inference Providers, using an open-weight language model for AI-powered study assistance.
Document processing: Apache PDFBox for extracting text from PDF study materials.
Storage: Browser local storage for keeping supported study data and progress without requiring a traditional database.
Deployment: Docker and Render for making the application accessible online.

I kept the application modular so that the frontend handles the study experience while Spring Boot manages document processing and AI requests.

One important design decision was to keep the Hugging Face API token on the backend rather than exposing it in the browser. The frontend communicates with Spring Boot, and the backend makes authenticated requests to the model provider.

The AI features are intended to help learners understand and practice their material. Generated answers and quizzes should still be checked against the original notes, because language models can make mistakes.

Why Does Open Innovation Matter?

More people can experiment with AI.

I wanted to build something beyond a basic chatbot. Open-weight models and accessible inference tools made it possible to experiment with AI-powered learning features without training a language model from scratch.

Developers have more control over their tools.

Using a model through a configurable backend makes it easier to experiment with alternatives as model availability, performance, and project requirements change.

Learning tools should be approachable.

Students should be able to experiment with AI features without needing a complicated setup or a paid subscription to multiple productivity tools. Study Buddy brings several study workflows together in one interface.

Open projects help others learn and improve.

By sharing the code, I hope other developers can explore the implementation, suggest improvements, add features, and adapt the project to their own learning workflows.

For me, open innovation is not just about making code public. It is about making experimentation more accessible and giving people the freedom to understand, adapt, and improve the tools they use.

What I Learned

Building Study Buddy helped me connect several parts of full-stack development: React, Spring Boot, document processing, API integration, environment configuration, and deployment.

It also reminded me that adding AI to an application is only one part of the challenge. The user experience, reliable API integration, clear limitations, and useful everyday workflows matter just as much.

What's Next?

I'd like to improve Study Buddy with more personalized revision recommendations, better tracking of weak topics, and additional ways to practice concepts from uploaded material.

The long-term goal is simple: make it easier for students to turn their notes into a practical learning routine.

Prize Categories

Open-Source AI / Best Use of AI

Study Buddy uses a language model through Hugging Face Inference Providers to support its study workflows. Rather than building a model from scratch, I focused on designing an application around AI: connecting it to study material, integrating it into a full-stack system, and making the results useful through a simple interface.

This project reflects what I find exciting about open innovation: developers can build practical applications around existing AI models, experiment with different options, and share their work with the community.

Thanks for checking out Study Buddy! 🎓💻

Study Buddy: a retro AI study companion built for a friend

Somesh Kumar Singh — Mon, 05 Oct 2026 01:21:27 +0000

devchallenge #weekendchallenge #hf26challenge #opensource

Hacktoberfest Weekend Challenge: Build for a Friend Submission 🤝
WebPage Link: https://study-buddy-u5vl-git-main-someshsingh2206-9453s-projects.vercel.app/

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

What I Built

Instead of switching between different tools for reading notes, making quizzes, and planning study sessions, Study Buddy brings these tasks together in one place.

I built the interface with a nostalgic retro aesthetic because studying doesn't have to feel like staring at another boring productivity dashboard.

I also wanted the project to be accessible and easy to try, without requiring users to create an account or set up a database just to get started.

Code

View Study Buddy on GitHub:https://github.com/Somesh-coding/study-buddy

Replace the link above with your actual public repository before publishing.

How I Built It

Study Buddy combines a Java backend with a modern React frontend.

I kept the application modular so that the frontend handles the study experience while Spring Boot manages document processing and AI requests.

The AI features are intended to help learners understand and practise their material. Generated answers and quizzes should still be checked against the original notes, because language models can make mistakes.

Why Does Open Innovation Matter?

More people can experiment with AI.

Developers have more control over their tools.

Using a model through a configurable backend makes it easier to experiment with alternatives as model availability, performance, and project requirements change.

Learning tools should be approachable.

Open projects help others learn and improve.

By sharing the code, I hope other developers can explore the implementation, suggest improvements, add features, and adapt the project to their own learning workflows.

For me, open innovation is not just about making code public. It is about making experimentation more accessible and giving people the freedom to understand, adapt, and improve the tools they use.

What I Learned

Building Study Buddy helped me connect several parts of full-stack development: React, Spring Boot, document processing, API integration, environment configuration, and deployment.

What's Next?

I'd like to improve Study Buddy with more personalized revision recommendations, better tracking of weak topics, and additional ways to practise concepts from uploaded material.

The long-term goal is simple: make it easier for students to turn their notes into a practical learning routine.

Prize Categories

Open-Source AI / Best Use of AI

Thanks for checking out Study Buddy! 🎓💻