StudySquad

Inspiration

Oftentimes it is difficult to keep track of everything going on in your classes, so we wanted to create an agent that could easily summarize anything from notes, questions to lectures.

What it does

Squad transforms how students engage with their coursework by creating a personalized AI squad for each class. Rather than juggling multiple platforms and generic AI assistants, students get intelligent, course-specific agents trained on their actual class materials. No more hunting through lecture notes or getting irrelevant ChatGPT answers. Just accurate, grounded responses sourced directly from your course materials.

How we built it

The Brains: Hybrid RAG Models We developed retrieval-augmented generation (RAG) models that automatically train themselves from professor-uploaded course content. Unlike generic LLMs that hallucinate or provide off-topic information, our RAG agents ground every answer in actual course materials—lecture notes, syllabus documents, Jupyter notebooks, datasets, and more. We implemented a sophisticated hybrid search algorithm that combines vector similarity search (semantic understanding) with full-text search (keyword precision), ranked through reciprocal rank fusion and cross-encoder reranking. This ensures students get the most relevant answers, not just the most confident ones.

Scalable RAG APIs: Our RAG endpoints are fully RESTful and horizontally scalable, designed to handle growing student loads without bottlenecks. Each course can maintain its own specialized agent while sharing underlying infrastructure

Multi-Source Indexing: Our FAISS vector store indexes PDFs, Jupyter notebooks, CSV datasets, and text files simultaneously, enabling unified search across diverse course materials

Key Features Course-Specific Intelligence - RAG agents trained only on your actual class materials Multi-Source Search - Simultaneously searches PDFs, notebooks, datasets, and lecture notes Honest Answers - Admits what it doesn't know instead of hallucinating Centralized Management - One dashboard for all your classes and deadlines Serverless Simplicity - No infrastructure headaches for day-to-day operations Enterprise-Ready - Horizontally scalable design ready for institution-wide deployment

Challenges we ran into

Working with large-scale Terraform files creates a lot of overhead considerations for how you want to orchestrate interactions between services. Also, training a model on the fly to guarantee the model always has up-to-date info.

Accomplishments that we're proud of

I am proud for parsing all the required file type- Jupiter notebook, tetx file, pdf & as well csv which are mostly the course material format lectures & prof's use. Created a large scale modular Terraform file for handling our cloud infrastructure needs, including file store, agents, databases, serverless computing, and API gateways.

What we learned

How to integrate and automate the development of RAG models for use in the classroom, and how we can use AI to assist those who need it the most.

What's next for StudySquad

The goal is to set up the service for professors so they can better manage their classes, and allow fine-grained control over what their agent has access to.

Built With

• amazon-web-services
• dynamodb
• fastapi
• langchain
• python
• rag
• react
• typescript
• vite