tResearch (tree-search)

Inspiration

As a group of researchers and learners constantly diving into academic papers, we wanted to streamline the process of discovering and connecting relevant research. Instead of manually combing through databases, we envisioned a smarter way to visualize and explore the relationships between papers — accelerating the path from question to understanding.

What it does

tResearch intelligently scrapes the web for scholarly resources related to a user’s query and visualizes them in a dynamic, tree-like structure. Each node represents a paper or resource connected to the user’s topic. Users can expand branches to explore related works and interact with a RAG-powered chatbot to ask questions and summarize insights from any paper within their research tree.

How we built it

We built tResearch with Next.js for a polished, responsive frontend and Express.js for the backend. Our AI agent leverages Bright Data MCP for web scraping and the Google Gemini API for intelligent reasoning and summarization. The RAG system is powered by ChromaDB Cloud, which stores paper embeddings for efficient retrieval and contextual understanding.

Challenges we ran into

Our biggest obstacle wasn’t technical — it was logistical. Unreliable Wi-Fi at CalHacks 12.0 made collaboration and testing difficult. We often had to move between locations to find stable connections while balancing time at the venue for meals and teamwork. Despite these setbacks, we kept pushing forward.

Accomplishments that we’re proud of

We’re proud to have built a working MVP under challenging conditions and to have explored the potential of agentic AI in accelerating research workflows.

What we learned

We deepened our understanding of agentic AI workflows, RAG architectures, and full-stack development. We also gained valuable experience integrating multiple AI tools into a cohesive system.

What’s next for tResearch

Our next steps include improving documentation, optimizing scraping efficiency, and accelerating the AI’s fetching and reasoning process. We also plan to enhance the visualization interface and expand integrations with academic databases.