DEV Community: keerthisree-25

slowmate

keerthisree-25 — Thu, 29 Oct 2026 14:44:13 +0000

SlowMate AI: A Little Ghost That Takes Care of the Boring Stuff 👻

This is my submission for Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine.

What I Built

What if the AI living inside your computer could quietly take care of the little things you keep postponing?

That's the idea behind SlowMate AI — an open-source AI assistant inspired by the theme Ghost in the Machine.

We often imagine AI as something that needs constant attention: open a chat, write a prompt, wait for an answer, and repeat. I wanted to explore a different direction: an assistant designed around reducing friction in everyday tasks, so we can spend less time managing work and more time doing what matters.

The inspiration also comes from a simple philosophy: slow and steady is still progress. Not every task needs a complicated workflow, and not every useful AI application needs to depend on an expensive, closed API.

SlowMate AI is my exploration of what a more accessible, open-source AI assistant could look like.

Demo

GitHub repository: https://github.com/keerthisree-25/slowmate-ai
Live demo: To be added.
Video walkthrough: To be added.

Code

Explore the project here:

GitHub — SlowMate AI

The goal is to make the project open to exploration, modification, and contributions from other developers.

How I Built It

I approached SlowMate AI with a focus on open-source AI and the possibility of running intelligence locally instead of relying entirely on paid AI APIs.

The direction is to build an assistant that can use an open-weight model through local inference, with application logic that connects the model to useful everyday tasks.

The main ideas behind the architecture are:

Open-source AI: Explore models that developers can run and experiment with themselves.
Local-first thinking: Reduce dependence on external AI services wherever practical.
Useful automation: Focus on tasks where a small amount of intelligence can save repeated effort.
Developer accessibility: Keep the project approachable for people who want to learn, modify, and extend it.

The implementation is still being developed, so the final model, framework, and working features will be documented here once verified.

Why Does Open Innovation Matter?

Open innovation makes AI experimentation more accessible.

A closed API can be a convenient way to get started, but it can also introduce usage costs, provider dependency, and limits on how much developers can inspect or customize the underlying system.

Open-weight models and open-source tools give developers more room to experiment, learn how systems behave, and adapt them to their own needs. Local inference can also help keep data on a user's machine, depending on how the application is designed.

For a student developer like me, this matters because building AI should not be limited to people who can afford expensive infrastructure.

I want SlowMate AI to be an opportunity to learn by building, make AI more approachable, and leave room for other developers to contribute their own ideas.

My Agent Session

Agent session: To be added if available.

The Spirit of It 👻

The ghost in the machine doesn't have to be frightening.

Maybe it's the quiet intelligence working behind the scenes — the little helper that takes one repetitive task off your plate, gives you back a few minutes, and lets you focus on something more meaningful.

SlowMate AI is my attempt to explore that idea: technology that helps without demanding all our attention.

Sometimes, the most useful intelligence is the kind that makes life a little less complicated.

Prize Categories

To be updated after checking the challenge's eligible partner categories and confirming which ones this project qualifies for.

side hustle

keerthisree-25 — Sun, 18 Oct 2026 18:06:19 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle
https://github.com/keerthisree-25/side-hustle
SideHustle AI: A Privacy-First Open-Source AI Assistant for Freelancers

What I Built

SideHustle AI is an open-source, privacy-focused business assistant built for freelancers, independent creators, tutors, and small business owners.

Managing a side hustle involves more than doing the actual work. People also need to organize receipts, track expenses, create invoices, monitor payments, and understand their finances. SideHustle AI brings these tasks together in one application.

Key features

AI Receipt Scanner: Extracts information from PDF invoices and receipt images using text extraction, OCR, and local AI models.

Expense Tracker: Organizes income and expenses with searchable records, categories, and CSV import/export.

Invoice Generator: Creates professional PDF invoices with customizable business details, line items, and tax calculations.

Payment Reminders: Generates polite email drafts for following up on outstanding invoices.

AI Business Assistant: Answers questions about spending and business finances using locally stored data and verifiable financial context.

Financial Reports: Generates monthly profit-and-loss summaries and exports reports as PDF or CSV.

Duplicate Detection: Identifies duplicate uploads using SHA-256 hashes and matching transaction details.

Who would pay for it?

Freelancers, consultants, designers, developers, and small business owners who want simpler financial management could benefit from this tool.

The core application is free and open source under the MIT license. A possible future business model is a paid Pro tier for advanced features such as bulk receipt processing and multiple business profiles.

The goal is simple: help people spend less time managing paperwork and more time growing their businesses.

Demo

Live demo: [Add your deployed application URL]

Video walkthrough: [Add your demo video URL, if available]

Until a hosted demo is available, you can run SideHustle AI locally using the instructions in the GitHub repository.

Code

The complete source code is available on GitHub:

Repository: https://github.com/keerthisreebandlamudi/sidehustle

The project is released under the MIT License.

Technology stack:

Frontend: React, TypeScript, Vite, Tailwind CSS

Backend: Python, FastAPI, Pydantic

Database: SQLite and SQLAlchemy

Open-source AI: Ollama with supported models such as Qwen 2.5, Gemma 2, and Llama 3.2

Document processing: PyPDF, Tesseract OCR, and Pillow

PDF generation: ReportLab

How I Built It

I built SideHustle AI using a local-first architecture that combines traditional software engineering with open-weight AI models.

Local document processing

The application extracts text from digital PDFs and uses OCR for scanned documents and images. When Ollama is available, a local language model helps structure the extracted information into useful invoice fields.

Reliable financial calculations

I deliberately kept financial calculations in Python rather than relying on an AI model to calculate totals. Invoice amounts, taxes, expenses, and profit figures are computed by application code to reduce the risk of incorrect financial answers.

AI with verified financial context

The business assistant retrieves relevant information from the local SQLite database and uses that context to answer questions about spending and outstanding invoices.

When the local model is unavailable, deterministic parsing and fallback logic help maintain core functionality.

A complete web application

The React frontend communicates with a FastAPI backend. Users can manage transactions, generate invoices, review financial summaries, and interact with the assistant through a single interface.

Testing and reliability

The repository includes automated tests for financial calculations, invoice totals, duplicate detection, PDF generation, and data retrieval.

Why Does Open Innovation Matter?

Financial information is sensitive. Receipts and invoices can reveal client relationships, business expenses, and personal information.

Open innovation made it possible to build an AI-powered financial assistant without making a proprietary cloud AI service a mandatory part of the architecture.

Privacy: Local inference through Ollama allows supported AI processing to happen on the user's machine rather than requiring financial documents to be sent to an external AI API.

Lower operating costs: Local models avoid per-request cloud AI inference charges, although users still need suitable hardware and may incur electricity and setup costs.

Freedom to experiment: Open-weight models can be swapped or evaluated without redesigning the entire application around one proprietary provider.

Transparency and customization: Developers can inspect the source code, adapt the application to their workflows, and contribute improvements.

Resilience: Combining local AI with deterministic fallback logic helps the application continue performing supported tasks even when the AI runtime is unavailable.

For me, open innovation is about making useful AI more accessible while giving developers and users greater control over their tools and data.

My Agent Session

I used an AI coding assistant during development to help with implementation, debugging, and testing.

Agent session: Add your DevRelay session link or the appropriate challenge-supported session embed if you have one.

Prize Categories

Hacktoberfest Open-Source AI Challenge — Week 2: Side Hustle

Open-source AI and open-weight models

Privacy-first and local-first software

WildLens 🌱 — AI Nature Companion that Tells You to Put Your Phone Away

keerthisree-25 — Fri, 09 Oct 2026 06:09:06 +0000

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

🌿 What I Built

Most AI applications try to maximize the amount of time users spend interacting with AI.

WildLens does the exact opposite.

WildLens is an AI-powered nature exploration companion that encourages people to go outdoors by turning ordinary walks into small discovery missions. The core guiding principle of WildLens is simple:

"Use AI to help people explore the real world, then get them off the screen."

"The goal isn't more screen time. The goal is more world time."

How it works:

Snap & Analyze: Take a single photo of a leaf, flower, bird, tree, insect, or rock found outdoors.
Open-Source Vision AI Result: Receive a fast identification ("This looks like..."), confidence rating, short description, nature fact, region prevalence, and safety disclaimers for mushrooms/wild items.
Touch Grass Challenge: Get a real-world sensory observation challenge tailored to your discovery (e.g. "Look around and find two other leaves that have a different shape" or "Spend two minutes listening quietly for another bird call").
Phone Down Mode: Switch into a minimal screen-free interface with a calming 5-minute timer and Web Audio API forest synthesizer that explicitly tells you: "📵 Now put your phone away and explore."
Return Check-In & Journal: When you return from exploring, share a quick reflection and log your real-world experience into a personal, non-addictive Nature Journal.

📽️ Demo

Here is the intended flow when taking WildLens outdoors:

Open WildLens ➔ Go Outside ➔ Find Something ➔ Take 1 Photo ➔ AI Analyzes ➔ Receive Touch Grass Challenge ➔ Put Phone Away ➔ Explore ➔ Return Check-In

Main Screenshots & Features

Look Up. Look Around. Hero landing with 1-click sample presets for instant camera-less testing.
Open Vision AI Analysis Card: Detailed botanical/zoological overview with safety caution disclaimers whenever confidence is low or caution items (fungi, toxic foliage, wild berries) are detected.
Phone Down Mode: Screen-free 5-minute countdown timer with built-in ambient breeze and songbird audio.
Return Check-In Modal: Quick selection pills (Saw something, Heard something, Found something new) + reflection notes + celebratory nature confetti.
Nature Journal & Stats: Category filtering (Leaf, Flower, Tree, Bird, Insect, Rock, Plant, Fungus) and mindful outdoor metrics (no infinite feeds, likes, or streak traps!).

💻 Code

keerthisree-25 / wildLens

🌿 WildLens — AI Nature Exploration Companion

Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

🌱 AI helps you discover something outside — then tells you to put the phone down

📖 Philosophy & Story

Most AI applications try to maximize the amount of time users spend interacting with AI.

WildLens does the opposite.

It uses AI for a few seconds to help someone understand something they discovered outside — and then tells them to put the phone away.

"The goal isn't more screen time. The goal is more world time."

🌟 Main User Experience Flow

Open WildLens ➔ Go Outside ➔ Find Something Interesting ➔ Take 1 Photo ➔ AI Analyzes ➔ Receive Touch Grass Challenge ➔ Put Phone Away ➔ Explore ➔ Return Check-In

Snap & Analyze: Take a photo of a leaf, flower, bird, tree, insect, or rock.
Open-Source AI Result: Get species identification, confidence…

View on GitHub

Repository: github.com/keerthisree-25/wildLens

🛠️ How I Built It

WildLens is built as a complete full-stack web application:

Frontend: React, Vite, Tailwind CSS, Lucide Icons, Canvas Confetti, and Web Audio API for ambient nature sounds.
Backend: Node.js, Express, SQLite (better-sqlite3), and Multer for image uploads.
Open-Source Vision AI Engine: A modular multi-provider vision engine built from the ground up to prioritize open-weight models:
- Ollama Local Inference: Supports running llava, qwen2.5-vl, moondream, or florence2 directly on a laptop or device via OLLAMA_HOST=http://localhost:11434.
- Hugging Face Inference API: Integrates with open-weight models like Salesforce/blip-image-captioning or Qwen/Qwen2.5-VL-7B-Instruct.
- Embedded Heuristics Fallback: An out-of-the-box zero-dependency fallback engine that analyzes visual features, color profiles, and botanical heuristics so developers can run and test the app immediately without needing API keys or hardware setups.

The vision provider is developer-configurable via environment variables:

VISION_MODEL=llava
VISION_PROVIDER=auto

💡 Why Does Open Innovation Matter?

Open innovation is essential for WildLens and nature exploration for four main reasons:

Offline Backcountry Trail Readiness: By running open-weight models like LLaVA or Moondream via Ollama, WildLens can function on a laptop or mobile device deep in national parks and hiking trails without cell signal or internet connectivity.
Complete Data Privacy: Your location and outdoor exploration photos stay on your device or private server, keeping your personal nature observations off centralized proprietary AI servers.
Developer Freedom & Model Swappability: Developers can easily swap vision models (LLaVA, Qwen-VL, Moondream, Florence-2) or plug in fine-tuned local botanical/zoological models as computer vision advances.
Community Prompts & Observation Logic: The observation task generator and safety evaluation rules are open source, allowing the community to improve species identification prompts and safety guardrails.

🤖 My Agent Session

Here is the complete agent session transcript showing how WildLens was conceptualized, built, tested, and verified:

Building WildLens: Touch Grass Nature Exploration Companion with Open Vision AI

You

Build a complete, polished web application called WildLens for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass. Use open-source AI at its core to get people off the screen and into the real world. Allow taking/uploading photos of leaves, flowers, birds, trees, insects, or rocks, analyze with open-weight vision AI, generate a Touch Grass outdoor challenge, switch into Phone Down Mode, and track findings in a Nature Journal.

Agent

I will build WildLens using React + Vite + Tailwind CSS for the frontend, Node.js + Express + SQLite (better-sqlite3) for the backend, and a modular Open-Source Vision AI Engine supporting Ollama, Hugging Face, and embedded vision heuristics.

▸ scaffold_application Initializing server, database schema, vision AI provider, touch grass challen...

Input

Initializing server, database schema, vision AI provider, touch grass challenge generator, and responsive UI components.

Output

Successfully created WildLens application with full-stack Node.js backend, SQLite database, React Vite Tailwind frontend, and modular open-weight vision AI integration.

You

Test the full-stack flow and verify API endpoints.

Agent

Verified /api/health, /api/analyze, /api/discovery, /api/journal, /api/stats, and /api/model-info. All tests passed cleanly!

🏆 Prize Categories

Overall Hacktoberfest Open-Source AI Challenge: Week 1 Winner (#hf26challenge)