Inspiration
We wanted to solve the problem of spending too much time searching through long research papers. Our goal was to create a smart assistant that quickly finds and summarizes relevant information for students and researchers.
What it does
Takes user queries in natural language.
Searches through research papers stored in TiDB Cloud using semantic embeddings.
Summarizes the most relevant results with a Hugging Face model.
Returns a concise and meaningful answer.
How we built it
Backend: Flask for lightweight API and routes.
Database: TiDB Cloud with vector search for semantic similarity.
AI Models: Hugging Face Sentence Transformers for embeddings + FLAN-T5 for summarization.
Frontend: Simple HTML form for input and displaying answers.
Deployment: Local testing with Python, ready for cloud deployment.
Challenges we ran into
Setting up TiDB Cloud connection with SSL certificates.
Handling embeddings and storing them as JSON in the database.
API errors while calling Hugging Face inference endpoints.
Limited time to test with a bigger dataset.
Accomplishments that we're proud of
Successfully integrated TiDB Cloud with vector search.
Built a working research assistant pipeline end-to-end.
Learned how to combine open-source models with cloud databases.
Created a functional demo in a short time frame.
What we learned
How to use TiDB Cloud for vector search.
How to connect Hugging Face models with Flask apps.
Importance of preprocessing and embeddings for semantic search.
Practical experience in building an AI agent from scratch.
What's next for Smart Research Assistant
Add support for uploading custom research PDFs.
Improve summarization with larger models.
Build a more polished frontend UI.
Deploy on cloud (Heroku, AWS, or Vercel) for public access.
Expand dataset to cover multiple domains beyond AI/ML.
Log in or sign up for Devpost to join the conversation.