We’re looking to eventually upload hundreds of documents for an enterprise chatbot. We also have some local tooling to sanitize all the info from our company knowledge base to convert it into a format that AI can better understand (e.g. markdown instead of a powerpoint file), in an attempt to guide Pickaxe to chunk info more effectively.
Despite all this, We’re concerned about how Pickaxe would perform at scale with a lot of knowledge in its base. Indeed, we could simply test and find out if it struggles, but if anyone has some insight to share before we do so, it would be great if you could share.
I’m also wondering how much of this could be in Pickaxe’s control. This video does a good job at explaining what’s going on in the backend at Pickaxe with RAG, but it still leaves some questions to be answered about scalability. Can we expect improvements the knowledge base performance and capability as LLMs advance? Currently we are limited to OpenAI models.
Thanks for sharing your concerns about scaling your enterprise chatbot with Pickaxe. Here’s a practical look at how you can enhance performance by integrating Pinecone, a robust vector database solution specifically designed for large-scale, AI-driven projects:
Scalability & Performance: While Pickaxe excels in user interface and lightweight AI processing, using Pinecone as your backend can significantly enhance scalability, providing efficient semantic search and faster retrieval even with extensive document libraries.
Security & Reliability: Pinecone offers enterprise-grade security (SOC2 Type II, HIPAA compliance, encryption at rest and in transit) and proven reliability for mission-critical applications—critical for securely managing sensitive enterprise data.
Optimized Data Retrieval: Pinecone is specifically optimized for handling high-dimensional vector data, enabling your chatbot to efficiently process and retrieve relevant information from large datasets. This makes it ideal for applications needing quick and accurate responses at scale.
Ease of Integration: Pickaxe supports seamless integration with Pinecone through its connected actions, letting you retain Pickaxe’s user-friendly UI while leveraging Pinecone’s advanced backend capabilities.
Futureproofing: As LLM technology advances, vector databases like Pinecone are built to evolve alongside these innovations, ensuring your knowledge base remains performant as newer models become available.
Given these advantages, integrating Pinecone into your Pickaxe setup is a strategic move, particularly if you’re planning on scaling to hundreds or even thousands of documents. This setup will not only enhance your AI’s responsiveness but also position your enterprise for smooth scaling as your knowledge base grows.
Let me know if you have any questions or need help getting started!
@user11 you can hit the Pinecone assistant endpoint from Pickaxe without using Make to retrieve an answer.
I’m using Make to watch the updates in a database (e.g. Airtable, Google Drive, etc…) and automatically update the Pinecone vector database. If you are a coder you can write a python program instead of using Make.This however is a simple no-code solution if you don’t want to use python.
Thanks for that insight. A tangentially related question, though. Any reason you hook up pinecone assistant to pickaxe as opposed to pinecone vector db?
It seems like pinecone assistant can do a lot of what pickaxe already does minus the frontend. I question the use of Pickaxe aside from that perk.
@user11 if you need help setting this up, one of my developers has time to look into your project. I can offer you a video meeting to check out your project. Just send me a DM for details. We’re happy to help!
