DEV Community: Vani thakur

I Built a Lazy Summarizer on a Tablet with Open-Source Gemma 4

Vani thakur — Fri, 23 Oct 2026 08:21:59 +0000

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

What I Built

I built a lazy text summarizer. You give it a long piece of text and it writes a two-line summary, so you don't have to read the whole thing. Anyone who gets long messages or articles can be a little lazier because of it. I built it on a tablet, with no laptop.

Demo

No deployed link, since I ran it in Google Colab. Here are screenshots of the model summarizing text:

Code

I ran the summarizer in a Google Colab notebook. The code is in the screenshots above, and the model is google/gemma-4-E2B-it from Hugging Face.

How I Built It

I used Gemma 4 (E2B-it), an open-weight model from Google, running in Google Colab on a free T4 GPU, all from my tablet's browser. I loaded it with Hugging Face Transformers and gave it text to summarize in two short lines.

Why open innovation mattered: Gemma 4 uses the Apache 2.0 license, so I could download and run it for free, without asking anyone for access. I first tried an older gated Gemma model and it blocked me with a 403 error until I accepted a license. The open one worked right away. Because the weights are open, I can also swap models or fine-tune it later.

Why Does Open Innovation Matter?

Open innovation matters because it allows you to access powerful, open-weights models like Gemma, giving you the freedom to build and experiment without the restrictions of closed systems.

My Agent Session

The Lazy Payoff

Prize Categories

Gemma

Global Travel Buddy — Your Complete AI Assistant for Traveling Anywhere

Vani thakur — Wed, 14 Oct 2026 15:10:41 +0000

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

What I Built

I built an AI travel assistant using Backboard AI that helps travelers anywhere in the world. It translates signs, tickets, and directions in any language, and gives information on food recommendations, safety tips, travel routes, and local culture and heritage. It would be useful for solo travelers, backpackers, and frequent international travelers who need quick, reliable help in unfamiliar places.

Demo

Here's a real test I ran with my assistant — I asked it a complex question covering language translation, bus routes, food recommendations, safety tips, hotel suggestions, and cultural information all at once, while roleplaying being in Hoi An, Vietnam covering language,food, safety and hotel follow up test same assistant handing queries about Himachal Pradesh and Korea .Here's how it responded:

Note: the assistant responded in Hindi and Chinese as well, showing its multilingual support for travelers from different backgrounds", "As you can see, the assistant accurately handled transport, food..."

Code

This project was built using Backboard AI's platform directly, so there isn't a separate GitHub repository — the prompt and configuration shown above represents the core of how the assistant

How I Built It

I used an open-weight AI model through Backboard AI's platform, which provides the agent framework and lets you configure a custom system prompt. I wrote a detailed instruction set that turns the model into a complete travel assistant — handling language detection and translation, transport route explanations, food and safety guidance, hotel recommendations, and cultural information, all built around that open-weight model powering the responses.

Why Does Open Innovation Matter?

Using an open-weight model meant I could fully customize how the assistant behaves instead of being limited by a closed API's rules. It also keeps costs low since I'm not paying per-request to a closed provider, which matters because this could actually work as an affordable side hustle for frequent travelers. And since everything runs on an open setup, I'm not locked into one company's pricing or features if I want to expand it later

Prize Categories

Best use of Blackboard

Running Workout Plan Generator — A Personal Fitness Coach Built in Python

Vani thakur — Sun, 11 Oct 2026 08:24:45 +0000

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

What I Built

I built a simple Python program that asks the user for their fitness level and how many days a week they want to run, then generates a personalized weekly running plan. It gives people an easy, approachable first step to get off their screens and start running, especially those who don't know where to begin.

Demo

I tested this locally in Google Colab. The program asks for the user's fitness level and preferred number of running days, then prints a complete personalized weekly workout plan.

Code

-- coding: utf-8 --

"""Untitled3.ipynb

Automatically generated by Colab.

Original file is located at
https://colab.research.google.com/drive/1WVPfMt0uvWrAO-ykSecu9EwaCDK-tR48
"""

fitness_level = input("What is your fitness level? (beginner / intermediate / advanced): ")
days_per_week = int(input("How many days per week do you want to run? "))
plans = {"beginner": "20 min light jogging + 5 min walk", "intermediate": "35 min run + 2 interval sprints", "advanced": "50 min run + hill training"}
print(f"\nYour weekly workout plan ({days_per_week} days/week):")
for day in range(1, days_per_week + 1):
print(f"Day {day}: {plans.get(fitness_level, plans['beginner'])}")

How I Built It

I used Python's built-in input function to collect user responses, and a dictionary to map fitness levels to suitable running plans. The whole project was built and tested directly in Google Colab, without needing any local setup.

Why Does Open Innovation Matter?

Open and lightweight tools like this make it possible for anyone, even without a gym or fancy equipment, to get a personalized plan and start moving. It lowers the barrier to going outside and being active.

My Agent Session

Prize Categories

I'm entering as an individual, in the general Touch Grass challenge category.

New to community

Vani thakur — Tue, 06 Oct 2026 08:44:35 +0000

Hello everyone...