Ideanet

Jatin Batchu — Thu, 08 Oct 2026 10:35:06 +0000

The problem

At every hackathon, someone builds something that already exists. Past projects are scattered across Devpost pages, GitHub repos and slide decks, so nobody can easily answer the simplest question before they start: has this been done?

We built Ideanet at Hacktoberfest Hack Day Coimbatore to answer it.

What we built

Ideanet is a social platform for hackathon builders. Hackathons are accounts, ongoing ones show up as stories, and every post is a project idea.

The part we care most about is Explore. You type an idea in plain language and get:

an originality score out of 10, with a short verdict
the closest existing projects, played like Instagram stories so you can tap through them
for each match, what is already built and how you could improve on it

Around that core we built:

sign up, log in, and a profile where you edit your bio and change your password
a For you feed of project ideas with comments
project pages
an Upcoming page for the next hackathons
light and dark themes, and a layout that works on a phone

How it works

The idea you type is turned into an embedding (nomic-embed-text) so it can be compared by meaning, not just by keywords.
We find the five most similar projects.
Gemma gets the idea plus those five projects and is told to act as a strict judge of originality only. It returns a score, its reasoning, and what is new about the idea.
We blend the two signals: 60% the language model's judgement, 40% an embedding-distance score.

Everything runs locally through Ollama, so ideas never leave your machine.

If Ollama is not running, the app falls back to a simple word-matching score so the page still works instead of crashing.

We also chose our wording carefully. Ideanet says an idea is similar to other projects. It never says "plagiarised" or "patented", because it can't know that.

How we built it

Part	Tech
Backend	Python, Flask
Frontend	Jinja templates, plain CSS and a little vanilla JavaScript (no framework, no build step)
AI	Gemma and nomic-embed-text through Ollama
Storage	JSON files
Auth	Hashed passwords, signed session cookies

Four of us split the work: backend and storage, the interface, login and data, and the AI scorer.

We used Claude as a coding assistant for parts of the frontend and for some debugging. We reviewed and tested what it produced, and the AI scoring itself is our own pipeline.

What we learned

Showing why an idea is not original (the matching projects) is more useful than a bare number.
A fallback matters. Local models can be off at demo time, so the app has to survive that.
Agreeing on one stack early would have saved us time. We spent part of the afternoon reconciling two different backend approaches.

What's next

Saving likes and comments permanently for every user
Posting your own projects from the app
Real chats between builders
Hosting it so the scorer is available without running Ollama yourself

Links

Built by Team Tesseract Testers: Jatin Naga Sai Batchu, Lakshanya Ilan Sezhiyan, Akshitha Venkatesh Devithulasimani and Aakaash V.

DEV Community: Jatin Batchu