PrakritiNaam: Connecting Local Nature Names with Open-Source AI

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

What I Built

PrakritiNaam (प्रकृतिनाम) is an open-source AI application that connects local-language names for plants and birds with their scientific identities.

In India, much of our knowledge about nature is passed down through families and local communities. We know plants by names like Tulsi, Kadulimb, and Pimpal, but connecting these names to scientific taxonomy isn't always easy.

I wanted to build something that helps people explore nature using the names they already know.

PrakritiNaam has two main features:

• What is it? — Search for plants or birds using local names in Marathi or Hindi. The application uses multi-sample consensus scoring to reduce unreliable predictions.
• Teach it a name — Add a local name, select a language, and write a short field note. A locally running Gemma model converts the note into structured JSON while preserving the name and language provided by the user.

The idea is simple: use AI to help us reconnect with nature and preserve the knowledge our communities already have.

Demo

Local application: Runs through FastAPI at http://127.0.0.1:8000.

Here is a look at PrakritiNaam in action! 🌿

Code

🔗 GitHub repository: https://github.com/kritikamandale/PrakritiNaam

How I Built It

PrakritiNaam combines open-weight AI, model experimentation, and a lightweight web application.

1. Species identification with Tinker

I used Tinker by Thinking Machines for fine-tuned species-identification experiments. The application samples five predictions and requires at least 60% agreement before accepting a result. When predictions are uncertain, it can indicate that verification is needed rather than blindly trusting the model.

2. Field-note structuring with Gemma

For the Teach it a name feature, I integrated Google's Gemma model using Hugging Face Transformers and PyTorch. The intended model is Gemma 4 E4B Instruct, running from local model files.

It helps transform everyday observations into structured records, while the backend preserves the user's original local name and language.

3. An offline-first approach

PrakritiNaam includes a seed taxonomy dataset and fallback logic for supported lookups when AI inference is unavailable. Its local model setup is designed to avoid dependence on a remote AI API for field-note processing.

4. Simple web interface

I built the backend using FastAPI and Uvicorn, with HTML, CSS, and JavaScript for a clean, nature-inspired interface.

Why Does Open Innovation Matter?

For me, open innovation means having the freedom to experiment, understand model behaviour, and build tools around real community needs.

Open-weight models make local inference possible, which can be especially useful in areas with unreliable internet connectivity. Processing field notes locally can also reduce the need to send community observations to external AI services.

Using tools like Hugging Face Transformers and Tinker lets developers experiment with fine-tuning and evaluate how well their models perform.

Most importantly, PrakritiNaam is built around the idea that technology should support local knowledge, not replace it.

AI can help connect names to scientific identities, but it should not pretend to know everything. Local experts and community knowledge remain essential for verification.

Prize Categories

• Touch Grass: Encourages outdoor exploration and learning about local plants and birds.
• Tinker by Thinking Machines: Uses Tinker for fine-tuned identification experiments and consensus scoring.
• Open-Source AI / Local Inference: Integrates Gemma through Hugging Face Transformers and PyTorch.

Building PrakritiNaam reminded me that AI doesn't always need to create something entirely new. Sometimes, it can help us reconnect with knowledge that has existed for generations.

Go outside, learn a local name for a plant, ask questions, and keep exploring. Let's use AI to bring us closer to nature. 🌿

Thanks for reading!