Inspiration
My inspiration and encouragement to initiate and finish this project was when I lost a hackathon and could not finish creating a RAG system that was going to assist rural children in availing themselves of personalized education. The initial idea was to produce quizzes and learning materials based on the strengths and weaknesses of individual students. That experience compelled me to come up with a more comprehensive Hybrid RAG system that could be implemented in educational as well as business settings.
What it does
A dual-retrieval system combining vector and graph-based search for smarter, context-aware responses. Built for education and enterprise use.
How we built it
- Vector DB: Titan embedding to convert to vector db to store as FAISS
- Graph DB: Neo4j + AWS Bedrock
- Parallel retrieval & smart prompt synthesis
- Personalized, scalable insights
Challenges we ran into
- First tried using normal OPEN AI API key due to rate limits → Switched to AWS Bedrock
- Complex integration & response synthesis
- Integrating proper CYPHER query to enhance Graph RAG
Accomplishments that we're proud of
- Outperforms traditional RAGs
- Handles ambiguity & complex relations
- Scalable architecture on AWS
What we learned
- Hybrid > Single-method RAG
- Vectors + Graphs = Best of both worlds
- Prompting & parallelism are key
- Query enhancement improves the retrieval
What's next for Hybrid RAG with AWS and Langchain
- Multi-modal support (text, image, tables)
- Live decision-making & streaming input
- Return to original vision: adaptive quizzes for students
Built With
- amazon-web-services
- bedrock
- embeddings
- langchain
- titan
Log in or sign up for Devpost to join the conversation.