PankhiPath: An Offline-First Bird Companion for Touch Grass

Prashant Mehta — Sun, 11 Oct 2026 12:10:28 +0000

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

🌿 PankhiPath: An Open-Source AI Bird Companion for Touch Grass

What if AI helped us spend less time looking at our phones and more time discovering the world around us?

That is the idea behind PankhiPath, a nature-focused web application designed to make birdwatching and outdoor exploration more accessible.

Instead of giving people another reason to stay glued to a screen, PankhiPath aims to make technology the starting point for a real-world nature experience.

🌱 What I Built

PankhiPath brings together bird discovery, outdoor planning, and open-weight AI in one application.

The main features include:

🐦 AI Bird Photo Identification: Upload a bird photo and get candidate matches using a locally running vision model.
🎙️ Bird Sound Identification: An experimental workflow using BirdNET to explore bird-call identification.
🌳 Nearby Parks: Discover nearby gardens and parks to find places for outdoor exploration.
🥾 Smart Nature Walk Planner: Plan a short walk with activities that encourage you to observe and listen to nature.
📵 Phone-Down Mode: Use a timer to encourage you to put your phone away and pay attention to your surroundings.
📚 Bird Guide: Explore bird information and learn about different species.
📔 Nature Passport: Save sightings and track your nature discoveries.
🇮🇳 Gujarati and English Interface: Make the experience more accessible to local users.

The guiding principle is simple:

Less scrolling. More noticing.

🎬 Project Demo

Watch the short project demo:

Video Demo: https://youtu.be/2-in7FIBhNw

🌐 Live Website: https://pankhipath.vercel.app

💻 GitHub Repository: https://github.com/prashantmehta1207-netizen/PankhiPath

The frontend is deployed on Vercel. Photo identification works in my local development environment, while the connection between the public website and the AI backend is still being tested.

The video is a feature overview with illustrative visuals, not a live end-to-end screen recording. The public website should therefore be treated as a UI preview rather than a reliably available online AI service.

🛠️ How I Built It

The project uses the following technologies:

React and Vite for the frontend.
Python and FastAPI for the local AI backend.
Hugging Face Transformers to load the image model.
OpenAI CLIP (openai/clip-vit-base-patch32) for local image-text matching.
BirdNET for experimental bird-sound identification.
Vercel for frontend deployment.
Cloudflare Quick Tunnel for testing remote access to the local backend.

For photo identification, the prototype compares an uploaded image against candidate bird labels and returns possible matches.

It is important to note that this is not a fine-tuned bird-species classifier. The highest-scoring match is a suggestion, not a guaranteed identification, and its score should not be interpreted as a calibrated probability.

🔓 Why Does Open Innovation Matter?

For me, open innovation means having greater control over how AI works, where data goes, and how a project can evolve.

1. Local inference and privacy

When I run the photo-identification model locally, the image can be processed on my own computer instead of being sent to a third-party hosted AI API.

This gives me more control over personal data and the ability to experiment with local inference.

2. No per-image AI API bill

Local inference avoids a separate hosted AI provider's per-request inference charge.

It still requires suitable hardware, storage, electricity, and an initial model download, but it makes experimentation possible without paying for every image request.

3. Freedom to experiment

Using an open-weight model lets me inspect the workflow, experiment with candidate labels, and explore alternative models instead of depending entirely on a closed API.

I can change the implementation and investigate other approaches as the project develops.

4. A more accessible nature tool

A local-first approach can be useful for nature enthusiasts who want to explore AI without depending on a paid inference service.

After the required model files have been downloaded, local photo analysis can run without sending every image to an external AI provider.

🌍 How PankhiPath Fits the Touch Grass Theme

AI is useful here only if it helps people engage with the real world.

PankhiPath is designed around a simple outdoor workflow:

Choose a nearby garden or park.
Plan a short nature walk.
Observe birds and listen to their calls.
Use AI to explore a possible identification.
Save the discovery and put the phone away.

The goal is to keep the screen interaction short and make the outdoor experience the main event.

🚧 Current Limitations and Next Steps

PankhiPath is a working prototype, and there is still work to do.

My next steps are to:

Finish and validate the public frontend-to-backend connection.
Test bird identification with more real-world photographs.
Validate the bird-sound identification workflow.
Take PankhiPath on an actual birdwatching walk and document its successes and mistakes.
Improve species matching and communicate uncertainty more clearly.
Improve reliability and make the installation process easier for other developers.

I have not yet completed a documented outdoor field test, so I will not claim real-world results that I have not verified.

💡 What I Learned

Building PankhiPath has taught me that making an AI model run is only one part of building a useful AI application.

The user experience, data privacy, model limitations, deployment, and reliability matter just as much as the model itself.

It has also helped me understand the difference between getting AI to work locally and making an AI-powered application reliably available to other people.

I want PankhiPath to be a small step toward using AI to reconnect people with nature rather than giving them another reason to keep scrolling.

🌿 Less scrolling. More noticing.

Feedback, suggestions, and contributions are welcome!

DEV Community: Prashant Mehta