what are you trying to build with TypeDB
what are you trying to build with TypeDB @cam ? if you don't mind sharing
2 Replies
hey @haikal , happy to share
@Joshua has been a huge help already, and @callum has been great too 🙂
a bit about us:
- we are leveraging LLMs to build self-improving software
- - this involves storing and re-writing versioned representations of code (ASTs, code property graphs, etc.) and code executions (logs, errors, data flows through code) in a graph structure
- - we've prototyped in RDF and relational structures but looking for more expressive constraints and polymorphic properties
- since the our knowledge graph will be maintained and improved by an LLM over time, schema flexibility with temporal versioning is important to us
- intending to model everything as close to 6NF as possible to give us the most flexibility, but maybe there are other "most flexible" approaches enabled exclusively by TypeDB,
- - because of the above, we'll likely be storing data immutably
- since our TypeDB graph will be storing code, it'd be cool to leverage store queries as data (Datalog has something akin to this - definitely not a necessity, but seems like it would mesh well with TypeQL)
- a subset our system will serve as a Knowledge Graph
we were in despair last week because after ~3 months of intensive research (we were working on defining requirements and searching for solutions as we went), we began to think we might have to write our own db
but then OpenAI's Deep Research led us here, and it seems very promising
I definitely think we're on the fringe and primed to push the edges of what TypeDB is intended for, but if it works it'll be magical
there has been a ton of hyper-relevant research on leveraging graph data structures for neurosymbolic AI from the past ~18 months, but the pace has really picked up in the past few months and it is now very obvious that pairing LLMs with graphs unlocks emergent capabilities and new use cases - we're having a lot of fun pushing on this and hope TypeDB can help us go from toy ideas to an application performant enough to deploy at scale
And it doesn’t, we’ll make sure it works!