AskAstra – Your Secure GenAI Enterprise Copilot

Inspiration

The fact that employees frequently squander time searching through emails, policies, and documents for straightforward answers served as our inspiration. We reasoned that since GenAI can now answer complicated questions, why not develop an AI that fully comprehends your business? AskAstra, an enterprise copilot that enables you to "talk to your work," was born out of this.

What it does

AskAstra is a secure GenAI assistant that allows users to ask questions and get prompt answers from their internal documents. You only need to upload documents, like reports or PDFs, and AskAstra will use RAG to generate accurate, natural-language responses, saving you time and making knowledge easily accessible.

How we built it

We selected Streamlit for a fast and efficient chat interface. For a quick and effective chat interface in the backend, we chose Streamlit. We used LangChain to create a RAG (Retrieval-Augmented Generation) pipeline on the backend. After being uploaded and stored in Amazon S3, files are transformed into text, embedded using Amazon Titan Embeddings, and saved in FAISS. In order to produce contextual responses based on the uploaded data, we employed Claude from Amazon Bedrock as the language model.

Challenges we ran into

Reducing latency when querying large documents was one of the most difficult tasks. We also needed to improve retrieval accuracy and change prompts to reduce hallucinations from the LLM. It was also challenging to control token limits during embedding and guarantee proper document parsing for jumbled PDFs. Throughout, maintaining data security and privacy was crucial.

Accomplishments that we're proud of

We created a functional GenAI assistant that uses enterprise documents to comprehend and react to real-world queries. During the hackathon, we managed RAG pipelines, integrated Amazon Bedrock, and developed a seamless chat experience. We're pleased that AskAstra uses private data to provide prompt, pertinent responses, showcasing the benefits, security, and influence of GenAI.

What we learned

We discovered how to integrate GenAI pipelines with AWS tools like S3 and Bedrock. We looked at how to query a vector store, embed documents, and increase answer accuracy using prompt engineering. We also learned how to create scalable, secure GenAI applications that work in actual business settings.

What's next for AskAstra – Your Secure GenAI Enterprise Copilot

Role-based access control will be implemented, Excel and dashboard analysis will be supported, and AskAstra will be integrated with Slack and Teams. The solution will be deployed on AWS as a fully managed SaaS product, making it scalable, secure, and prepared for enterprise adoption. We also aim to improve accuracy through timely tuning and incorporate user feedback.