Sales Project Status Report

Inspiration

Emails are brimming with untapped, unstructured data, often left unrecorded in structured databases. Imagine the power of harnessing this data to automatically generate actionable insights! A key opportunity lies with Sales Managers and Directors, who can use this data to create reports and stay on top of their customer relationships effortlessly.

What it does

Our app takes any high-level query, breaks it down into sub-questions, and uses Retrieval-Augmented Generation (RAG) to pull relevant information. It then summarizes the data and compiles it into a concise, executive-style report in Markdown format—perfect for decision-makers.

How we built it

The stack is LlamaIndex Workflows and RAG with Pinecone, Pinecone for the Vector store, Arize Phoenix for tracing, Reflex for a UI, Toolhouse for querying the web and news sources, and OpenAI models and embeddings.

Challenges we ran into

Along the way, we encountered some hurdles—like identifying a reliable data source, configuring Pinecone and Arize Phoenix, and the time to extract and turn 1.4GB of emails into embeddings and uploading them to Pinecone. But these roadblocks only fueled our drive to succeed.

Accomplishments that we're proud of

Successfully integrating each tool including the LlamaIndex Workflows, with RAG, Pinecone, and also getting Arize setup with LlamaIndex. The final results are great as well.

What we learned

Working with new libraries involves reading a lot of documents. Cursor IDE can help if iteratively feeding error stack traces back to the code generation prompt to fix errors.

What's next

Evals, making sure the existing use cases work in real scenarios with real data, and iterating on the app to embed them into our customer's day to day work

Built With

• arize
• llamaindex
• openai
• pinecone
• python
• reflex
• toolhouse