https://github.com/vero-code/castanea

Multi-agent AI research assistant for students. Powered by Google ADK, Perplexity, and Gemini models.
https://github.com/vero-code/castanea

agents ai analysis automation edtech fastapi gemini google-adk hackathon perplexity rag react research vite writing

Last synced: about 1 month ago
JSON representation

Multi-agent AI research assistant for students. Powered by Google ADK, Perplexity, and Gemini models.

• Host: GitHub
• URL: https://github.com/vero-code/castanea
• Owner: vero-code
• License: mit
• Created: 2025-08-09T12:28:06.000Z (about 2 months ago)
• Default Branch: master
• Last Pushed: 2025-08-09T20:42:36.000Z (about 2 months ago)
• Last Synced: 2025-08-09T20:42:58.095Z (about 2 months ago)
• Topics: agents, ai, analysis, automation, edtech, fastapi, gemini, google-adk, hackathon, perplexity, rag, react, research, vite, writing
• Language: Python
• Homepage: https://devpost.com/software/castanea
• Size: 89.8 KB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Castanea

![React](https://img.shields.io/badge/React-61DAFB?style=for-the-badge&logo=react&logoColor=black)

![Python](https://img.shields.io/badge/Python-3776AB?style=for-the-badge&logo=python&logoColor=white)

![Google ADK](https://img.shields.io/badge/Google%20ADK-4285F4?style=for-the-badge&logo=google&logoColor=white)

![Perplexity API](https://img.shields.io/badge/Perplexity-000000?style=for-the-badge&logo=perplexity&logoColor=white)

> **Three smart AI agents that find, analyze, and create — so you can focus on what matters.**

Castanea is an AI-powered multi-agent platform for students and researchers. It automates time-consuming academic tasks by combining **real-time search**, **text analysis**, and **content generation** in one place. 

Built with the power of **Google ADK** and **Perplexity API**, it features three specialized AI agents coordinated to deliver accurate, well-structured, and useful results. Created for [PANDA Hacks 2025 🐼🚀](https://devpost.com/software/castanea)

## Features

Castanea operates through a team of specialized AI agents, each designed for a specific academic task:

-   ### 🧠 ResearcherAgent

    

    Performs in-depth research on complex topics. It leverages the **Perplexity** search engine to gather real-time, factual data and then uses the power of **Gemini Pro** to analyze, synthesize, and structure this information into a comprehensive and coherent report. This Retrieval-Augmented Generation (RAG) approach ensures answers are both accurate and well-written.

    

-   ### ⚡ AnalystAgent

    

    Provides quick and efficient analysis of existing text. Built on the **Gemini Flash** model for maximum speed, this agent can instantly summarize articles, extract key points from lecture notes, or identify the main arguments in a difficult text, helping students grasp core concepts faster.

    

-   ### ✍️ WriterAgent

    

    Acts as a versatile writing assistant. It can generate original content like essays, reports, and formal emails based on a user's prompt. Additionally, it is equipped with a tool to save its output directly to a file (e.g., social_media_essay.txt), making it easy to store and edit the work.

## Technologies

- Vite, React, JavaScript

- Python

- Agent Development Kit (ADK)

- Gemini Pro/Flash

- Perplexity’s Sonar API

## Installation

1. Clone the repository:

	```bash

	git clone https://github.com/vero-code/castanea.git

	cd castanea

	```

2. Configure the `.env` file:

	- Create `.env` in the root of the project.

	- See an example in `.env.example`.

3. Run the services:

	```bash

	# Frontend

	cd frontend

	npm install

	npm run dev

	```

 

	```bash

	# Backend

	cd backend

	python -m venv .venv

	.venv\Scripts\Activate.ps1

	pip install -r requirements.txt

	```

## Testing

### Test for ResearcherAgent  

The CRISPR and GMO problem is a complex, topical issue that requires gathering information from different fields (ethics, biology, law).

It is an ideal stress test for perplexity_search and the agent's ability to synthesize a complex answer.

```

Write a detailed report comparing the ethical implications of CRISPR gene editing with traditional GMO technology.

```

### Test for AnalystAgent

Analyzing a scientific abstract is something that students encounter all the time. 

The task checks whether the agent (Flash model) can quickly and accurately extract the essence from a scientific, "dry" text.

```

Summarize the key finding of this scientific abstract in a single, clear sentence:

This study investigates the correlation between bilingualism and delayed onset of dementia. 

A longitudinal study was conducted over 10 years with a cohort of 500 monolingual and 500 bilingual participants. 

Cognitive decline was measured using the Mini-Mental State Examination (MMSE) annually. 

Results indicated that the bilingual cohort exhibited first symptoms of dementia on average 4.5 years later 

than the monolingual cohort, suggesting that the constant cognitive effort of managing two languages 

builds cognitive reserve, thereby providing a protective effect against neurodegeneration.

```

### Test for WriterAgent  

Writing an argumentative essay is not just text generation, but a demonstration of the ability to build a logical structure (thesis, arguments, conclusion). 

Using the save_report tool here is absolutely logical - the student will want to save his work.

```

Write a 5-paragraph argumentative essay on whether social media has a net positive or negative impact on teenage mental health.

Save the output to 'social_media_essay.txt'.

```

## 📜 License

This project is licensed under the [MIT License](LICENSE).