#41 - Prospectus

Prospectus is an AI guide to educational institutions all over the world.

Comment

Inspiration

Every year, millions of students all over the globe face a life-changing decision: college admissions. The reasoning behind the creation of 'Prospectus' is rooted in its original meaning – 'to look forward.' Our product strives to help students see the path ahead to different colleges, one query at a time, towards a future brimming with educational possibilities.

What it does

Prospectus is a rich AI assistant designed to assist you with answering questions about specific colleges. Prospectus is designed to answer questions ranging from general Academics and Campus Life questions to even specific Meal Plans and Menus!

How we built it

We built Prospectus using RAG Pipelining and Processing techniques. We used React to build the Framework and Django for the backend functions. We used Azure OpenAI for the LLM and Llama Index for the Ingestion Framework as well as MongoDB Atlas for Vector Storage. For our actual implementation however, our NLP pipeline is a tad unorthodox. Instead of going for the recommended Azure AI Search approach, we decided to use GPT4 with custom embeddings by VoyageAI, which are known for their top performance on the Massive Text Embedding Benchmark.

Challenges we ran into

Despite the advantages of our unorthodox approach with custom vector embeddings and Ingestion frameworks are beneficial in terms of performance and cost efficiency, we ran into challenges trying to integrate everything together.

Accomplishments that we're proud of

Despite our challenges, we are proud to have a completed cloud infrastructure plan and a high-level architecture design. We also had the list of technology, that we narrowed down through research and experimentation, that would best suit our design. We also created several difficult components of the project within the limited timeframe, such as a full user-friendly website, the backend services incorporating custom vector embeddings models.

What we learned

We learned a lot about various frameworks and tools, such as Azure OpenAI, MongoDB Atlas, Llama-Index, etc. More importantly, we learned how these tools can integrate with each other to become end-to-end solutions. Several of us also got to experience high-level architecture design.

What's next

While the hackathon may be over, college admissions are not going anywhere. Our team would love to continue to working on our bot, which truly has the potential to become a cost efficient powerhouse in the Virtual Assistant market.

Built With

Share this project:

Updates