DEV Community: Niraj Nagrale

LocalLedger: Private Invoice Sorting for Freelancers

Niraj Nagrale — Tue, 13 Oct 2026 12:30:21 +0000

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

What I Built

LocalLedger is a local-first invoice inbox for freelancers. It turns invoice text into a reviewable table of vendor, dates, amounts, category, and payment status, then exports the results to CSV. It is aimed at solo freelancers who want to spend less time copying invoice details into a spreadsheet without sending client documents to a third-party AI API.

Demo

Run it locally by following the setup in the README. A public demo link or video can be added here when one is available.

Code

View the Challenge 3 source on GitHub.

How I Built It

The interface uses Streamlit. The app extracts text from text-based PDFs and text files, then asks the open-weight qwen2.5:3b model through a local Ollama instance to return structured invoice fields as JSON. Users can review the results and download a CSV. Invoice content is sent only to the user's local Ollama service; this version does not implement scanned-document OCR.

Why Does Open Innovation Matter?

Invoices contain sensitive business information. Running an open-weight model locally means invoice text does not have to go to a hosted model provider, and there is no per-document inference charge after downloading the model. Users can inspect or change the extraction prompt and swap to another Ollama-compatible model rather than being locked to one provider's API and pricing.

My First User

I have not yet tested this with a first user. This section should be updated after showing the app to a freelancer and recording their actual feedback.

Prize Categories

Main Hacktoberfest Open-Source AI Challenge: Side Hustle.

Touch Grass Planner: Local AI for Your Next Outdoor Reset

Niraj Nagrale — Tue, 06 Oct 2026 14:02:06 +0000

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

What I Built

I built Touch Grass Planner, a local AI app that helps people choose a real-world activity when they want to step away from the screen.

Instead of asking a closed API for recommendations, the app uses an open-weight sentence-transformer model running locally on the machine. The user enters a mood or goal such as “I want a calm walk with birds and fall colors,” and the app recommends a few outdoor options based on weather, time of day, energy level, and intent.

This is designed for people who want a small reset without planning a whole hiking trip: bird walks, garden sessions, sunrise strolls, social trail walks, and easy nature resets.

Demo

A live local version runs with Streamlit:

streamlit run app.py
Then open the local app in the browser at http://localhost:8501.

Code

GitHub repo: https://github.com/nirajnagrale7/hacktomberfest-second-challenge-2026.git

How I Built It

I built this with a simple local AI stack:

Streamlit for the app UI
Hugging Face sentence-transformers for the embedding model
PyTorch for on-device inference
A local JSON catalog of outdoor activities for matching user intent to real-world activities
The app uses the open-weight model to embed the user prompt and compare it against outdoor activity descriptions. It then ranks the best matches based on semantic similarity plus a few custom heuristics for weather, energy, time of day, and goal.

The important part is that the model runs on-device. There is no paid subscription, no remote API, and no external AI service required for the recommendation flow.

Why Does Open Innovation Matter?

Open innovation matters here because this project is built around privacy, affordability, and control.

The core experience works locally: the model runs on the user’s machine, the recommendation logic stays transparent, and the app can be customized or fine-tuned without being locked into a closed platform. That makes it more practical for real-world use in places where internet access may be unreliable or where a person wants to avoid sending personal preferences to a third-party server.

A closed model would have made this app harder to trust, harder to experiment with, and more expensive to operate. Using open-weight tools allowed me to build something that is lightweight, local, and adaptable — exactly the kind of setup that makes AI useful in everyday life instead of only in a cloud dashboard.

The project is intentionally small, but it shows how open-source AI can make an app more personal and more grounded in the real world.

My Agent Session

I used a local, open-source workflow to prototype the idea directly on-device and iterate on the recommendation logic without using a closed AI service.

Open Meal Buddy: A Local AI Meal Planner for Someone You Love

Niraj Nagrale — Fri, 02 Oct 2026 04:00:08 +0000

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

What I Built

I built Open Meal Buddy, a local AI meal-planning helper for my roommate.

It takes their allergies, pantry items, food preferences, and weekly budget and turns them into a practical meal plan without sending their personal data to a third-party server. The goal was to make healthy, affordable cooking easier for someone who is busy, has dietary restrictions, and wants something simple to follow.

Demo

A working local demo is available in the browser at:

http://127.0.0.1:5000

You can also run it from the terminal with:

python app.py --name "Ava" --allergies "peanuts,dairy" --budget "40" --days 3 --pantry "rice, eggs, spinach, chickpeas, onion" --likes "vegetarian, quick meals" --schedule "busy weekday dinners"

Code

https://github.com/nirajnagrale7/hacktoberfest-first-challenge-2026

How I Built It

I used Ollama with an open-weight local model and a lightweight Python app that builds a personalized meal plan from real-life constraints.

The project is built around open-source AI because the model runs on a laptop instead of in the cloud. The app structures the prompt with:

allergies and food restrictions
pantry ingredients already at home
a weekly budget
preferences like vegetarian, quick meals, or family favorites
schedule constraints such as busy weekday dinners This makes the system easy to run, easy to customize, and private by default.

Why Does Open Innovation Matter?

Open innovation matters here because it keeps the experience private, affordable, and under the user’s control.

Instead of sending a person’s dietary history and food preferences to a closed API, the app can run locally on a laptop. That means:

no personal food data is sent to a remote server
no API cost for each generated plan
easier model swapping and fine-tuning later
no dependency on internet access for day-to-day use A closed AI service could generate similar meal plans, but it wouldn’t be as private, cheap, or adaptable for a real friend-focused tool like this.

My Agent Session

This project was built as a practical local-first assistant for a real person and a real problem: making healthy meal planning easier without the stress of allergies, budget, and time constraints.

Prize Categories

Best use of open-source AI
Best project built for a friend or loved one

Hacktoberfest 2026 is here!

Niraj Nagrale — Thu, 01 Oct 2026 13:52:13 +0000

It’s that time of the year when developers around the world come together to contribute, learn, and grow through open source. 🚀

💻 Whether you’re fixing bugs, adding features, or improving documentation, every pull request counts.
🌍 Join the global community, sharpen your skills, and make an impact.
🎁 Plus, earn cool swag and bragging rights along the way!

Let’s code, collaborate, and celebrate open source this October. 🔥

Hacktoberfest #OpenSource #DevCommunity