DEV Community: Harsh Vyavahare

FuelLedger AI — Smart Fuel Station Intelligence Platform.

Harsh Vyavahare — Mon, 19 Oct 2026 05:45:07 +0000

Here is a complete, publication-ready submission post tailored for your FuelLedger AI project for the DEV.to Hacktoberfest Open-Source AI Challenge (Week 2: Side Hustle).

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

What I Built

FuelLedger AI is an offline-first, privacy-focused ERP and intelligent copilot engineered specifically for independent petrol and fuel station owners (with deep localization for Indian retail outlets — IOCL, BPCL, HPCL, and private dealers).

🎯 Who Would Pay For It?

Independent fuel station franchisees, fuel retail operators, and highway fuel outlet dealers managing multi-shift daily operations, fluctuating prices, and multiple payment rails.

🛑 The Problem It Solves

Running a modern fuel station is a high-volume, thin-margin, operationally chaotic business:

The Payment Reconciliation Nightmare: With over 65% of daily transactions moving to UPI and card POS machines, fuel pump owners lose thousands of rupees every week to uncredited payments, bank gateway reversals, and employee shift handover slips.
Inventory Stockout vs. Dead Capital: Fuel deliveries require advance capital. Under-ordering leads to dry pumps and lost regulatory licenses; over-ordering ties up tens of lakhs in working capital.
Data Privacy & SaaS Bill Fatigue: Station owners refuse to upload proprietary billing, shift handover, and tank dip numbers to unvetted cloud SaaS vendors with expensive monthly subscription fees.

💡 The Solution

FuelLedger AI provides an all-in-one local ledger and AI copilot:

Shift & Fuel Sales Logging: Real-time tracking of Petrol, High-Speed Diesel (HSD), and CNG dispensing units.
Smart 3-Way Reconciliation: Cross-checks physical dispenser meter readings vs. cash handovers vs. digital UPI / POS settlement reports to flag discrepancies instantly.
EV Charging Bay Integration: Future-proofed logging for electric vehicle charging infrastructure alongside conventional nozzles.
Conversational AI Co-Pilot: An on-device assistant that queries inventory trends, predicts reorder days, parses natural-language sales logs, and drafts transactions.

Code

Github Repo : https://github.com/Harshvyavahare11/FuelLedger

How I Built It

FuelLedger AI is built around a Local-First, Edge AI philosophy:

┌────────────────────────────────────────────────────────┐
│             FuelLedger AI (React 19 + Vite)            │
├────────────────────────────┬───────────────────────────┤
│   Local Business State     │    On-Device AI Layer     │
│  (Sales, Dips, Cash/UPI)   │     (aiService.js)        │
└─────────────┬──────────────┴─────────────┬─────────────┘
              │                            │
              ▼                            ▼
      Browser LocalStorage       Ollama Local Server
     (100% Offline Ledger)     (http://localhost:11434)
                                           │
                                           ▼
                                 Gemma 3:4B / Llama 3

Open-Weight Model Integration (gemma3:4b via Ollama):
- Rather than relying on cloud LLM APIs, the system directly interacts with http://localhost:11434/api/generate and /api/tags.
- Running Google's lightweight open-weight Gemma 3 (4B) allows station owners to run private inference on a standard retail counter laptop or modest edge desktop without high-end GPUs.
Domain-Specific Prompt Harnessing & Zero-Shot Intent Parsing:
- The AI assistant includes a custom system context configured for petroleum compliance, Indian currency formatting (₹ Lakhs/Crores), and shift dynamics.
- Includes Natural Language Transaction Parsing: An operator can type: "Logged 45L of diesel paid via UPI" and the AI extracts structured JSON drafts ({ fuelType: 'diesel', quantity: 45, paymentMethod: 'UPI' }) with confirmation guards.
Resilient Offline Architecture:
- If Ollama is not detected or the operator is in transit, aiService.js detects the connection drop within 3 seconds and seamlessly switches to an embedded demo knowledge engine, ensuring zero downtime.

Why Does Open Innovation Matter?

Open innovation and open-weight models transformed FuelLedger AI from an unviable idea into a practical, marketable side hustle:

Zero Recurring Marginal Cost: Fuel station margins are ultra-tight (typically ₹2 to ₹3 per litre). A SaaS model charging $50–$100/mo in cloud AI API token bills is a non-starter for tier-2/3 operators. Open-weight models like Gemma 3 make the marginal cost of intelligence $0.00.
Total Data Sovereignty & Offline Reliability: Fuel stations are often located on rural highways where broadband can be intermittent. Local AI ensures the assistant operates even during fiber cuts. More importantly, financial records, tank inventory numbers, and daily cash tallies stay completely confidential on the station's local machine.
No Vendor Lock-In: Because the inference interface adheres to open standards, station operators can freely swap models — from gemma3:4b to llama3.2:3b or quantized Mistral — as edge hardware evolves.

My Agent Session

Developing FuelLedger AI involved complex business logic: nozzle dip math, shift reconciliation matrices, and local REST bridges to Ollama.

Using AI agent pair programming allowed rapid scaffolding of domain-specific React components, Recharts visualizations, and prompt harness engineering.

Agent workflow and prompts used during development are archived in the repository's .system_generated project docs.

My First User

To test the system against real-world chaos, I shared an alpha build with Rajesh P., an independent fuel dealer managing an outlet on the outskirts of Nagpur:

"The reconciliation tab caught a ₹1,250 UPI transaction discrepancy on Day 2 that my shift boys had missed during shift change. I love that it doesn't need me to create an account or send my pump sales data to random cloud servers. If this comes packaged as a simple desktop installer with WhatsApp daily summaries, I'd easily pay a one-time setup fee of ₹15,000 to ₹25,000 for it."

Prize Categories

Primary Track: Open-Source AI Challenge Week 2: Side Hustle
Partner / Ecosystem Alignment: Open-Weight Edge Inference (Gemma / Llama via Ollama)

I built PawPlan WebApp- To spend time with pets and nature

Harsh Vyavahare — Sun, 11 Oct 2026 12:35:15 +0000

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

What I Built

🐾 PawPlan — From Screen Time to Green Time 🌿

POV: Your phone's screen time is 📈, but your pet's outdoor time is 📉. We need to fix that. 💀

Meet PawPlan, a pet-care and outdoor activity platform designed to get people off their screens and out into the real world — with their furry (or not-so-furry) besties. 🐶🐱

Let's be real: we spend hours doom-scrolling, while our pets are probably wondering, "Bro, are we going outside today or what?" 😭

PawPlan aims to change that.

🐕 Plan the walks: Create a weekly walking schedule and share pet-care responsibilities among family members.

🌳 Touch actual grass: Discover nearby parks and outdoor spots instead of spending the whole day indoors.

👟 Track the adventure: Record walking duration, distance, and routes when location tracking is available.

🤝 Find your pet people: Discover other pet owners with similar pets, connect with them, send friend requests, and chat. Who knows? Your dog's next bestie might introduce you to yours. 👀

💚 Build a community: Make outdoor time a social habit, not another task on your to-do list.

The goal isn't to make you spend more time on an app. It's to make you open the app, make a plan, put your phone away, and GO TOUCH GRASS. 🌿

Demo

🐾 Try PawPlan: https://harshvyavahare11.github.io/PawPlan/

Code

💻 GitHub Repository: https://github.com/Harshvyavahare11/PawPlan

Built in the open, with room to grow. Contributions and feedback are welcome!

How I Built It

PawPlan is being built as a modern, responsive web application using:

React + Vite — for the frontend and fast development.
Tailwind CSS — for a clean, mobile-friendly UI.
Lucide Icons — for lightweight, consistent interface icons.
OpenStreetMap — as the foundation for open geographic data and nearby outdoor discovery.
Browser Geolocation API — for location-aware features and walk tracking, with user permission.
GitHub Pages + GitHub Actions — for deployment.

The AI direction is to integrate an open-weight model for personalized walking plans and household scheduling, while keeping actual scheduling constraints and safety checks in application logic.

The goal is to make open AI useful in everyday life, not just add a chatbot because AI is trending. 👀

Note: The local-model integration and any simulated social features should be described as planned or demo-only until they're implemented and verified.

Why Does Open Innovation Matter?

Because your pet's routine shouldn't depend on an expensive AI subscription or handing over every little detail to a black-box service. 🫡

Open-weight models give developers the freedom to experiment, swap models, customize behavior, and explore local inference. Open mapping data also offers an alternative to building everything around a proprietary mapping ecosystem.

For PawPlan, the long-term vision is a local-first, privacy-conscious experience where household schedules and pet profiles can stay on the user's device wherever practical.

And imagine having your weekly plan available even when the park has zero signal. No Wi-Fi? No problem. Touch grass, not the refresh button. 🌱

Open innovation also means more people can inspect, adapt, and contribute to the project instead of being locked into one provider.

The real flex isn't using AI to keep people glued to their screens. It's using technology to help them put their phones down.

My Agent Session

I'll add my development session here once it's recorded and ready to share.

Prize Categories

Open-Source AI Challenge — Week 1: Touch Grass 🌿

ReadSpace- A personal book

Harsh Vyavahare — Sun, 04 Oct 2026 06:23:44 +0000

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

What I Built

I built ReadSpace, a modern, responsive personal book management and reading tracker for avid readers of all ages.

I designed and built ReadSpace for my friend, a passionate bookworm who reads both physical paperbacks and digital e-books. She always struggled with:

Losing track of physical ribbon bookmarks across multiple books she was reading simultaneously.
Forgetting memorable quotes and personal takeaways hidden inside scattered notebooks.
Feeling overwhelmed by strict reading goals without knowing their required daily reading velocity.

ReadSpace solves these problems by providing a distraction-free, privacy-focused library web app featuring:

Dashboard: Reading stats, streak counters, currently reading focus cards, annual book challenges, finish date countdowns, and recent log timelines.
My Library: Multi-shelf organization (Currently Reading, Want to Read, Completed, On Hold) with search, genre/format filters, sorting, and seamless Grid/List view toggling.
Reading Velocity Tracker: Automatic daily pace recommendation engine ("Read 24 pages/day to finish on schedule!").
My Notes & Quotes Journal: Centralized repository for favorite passages, takeaways, and page bookmarks with one-click copy-to-clipboard functionality.
Physical Book Mode: Distraction-free offline reading stopwatch timer and physical ribbon bookmark page synchronizer.
Discover Catalog: Curated book recommendations with custom shelf selection.
Privacy & Persistence: Complete LocalStorage persistence with Warm Library / Dark Mode aesthetics and JSON data backup/export.

Demo

Live Deployed App: https://harshvyavahare11.github.io/ReadSpace-Built-for-a-friend/

Code

GitHub Repository: https://github.com/Harshvyavahare11/ReadSpace-Built-for-a-friend

Local Preview: Run npm install and npm run dev to launch locally at http://localhost:5173/.

How I Built It

ReadSpace was built using a modern open-source web stack:

Frontend Core: React 18 + Vite
Styling & Aesthetics: Tailwind CSS v4 with custom warm off-white background palettes (#FBF9F4), deep navy typography (#1E293B), forest green accents (#2E5A44), book spine 3D gradient effects, and smooth micro-animations.
Icons & Visuals: Lucide React
Effects: Canvas Confetti for celebratory book completion triggers!
State & Persistence: Custom React Context API with automatic LocalStorage synchronization.

The project architecture was developed using AI pair-programming agent workflows to rapidly scaffold reusable components, calculate dynamic reading velocity math, and refine responsive UI layouts.

Why Does Open Innovation Matter?

Open innovation and open-source software empower users to take full control of their personal data. Unlike commercial reading tracker apps locked behind subscriptions, proprietary algorithms, or invasive data tracking:

Privacy & Data Ownership: ReadSpace keeps all reading data, notes, and habits stored locally in the browser. Users can export their entire library backup to a JSON file at any time.
Offline & Distraction-Free Freedom: Open design allows dedicated physical reading sessions without mandatory cloud logins or server downtime.
Customizability: Anyone in the community can fork the repository, add custom metadata fields, or connect custom APIs.

Prize Categories

Hacktoberfest Weekend Challenge: Build for a Friend
Open Innovation

Excited to start learning with dev

Harsh Vyavahare — Sat, 03 Oct 2026 13:39:18 +0000

I am Harsh Vyavahare . I am new to dev community