9 Comments
User's avatar
Denis Craciun's avatar

Amazing article. I’ll definitely implement this into a real scenario in the next weeks. Thank you :)

Paul Iusztin's avatar

My pleasure! Ping me with what you built. Would love to take a look

Tap's avatar

I wonder, Is RLM a harnesses agent ?

and will the harness agent replace RAG pipeline ? (chunk,embed, retrieve)

Paul Iusztin's avatar

Yes, you can think of RLM as an orchestration technique used in harnesses.

Depending on your app, it can replace RAG or complement it.

What are you thinking of?

Alex Openstone's avatar

This is cool, Paul. I think works well for regulated type industries (healthcare, pharma) where many documents are just not well indexed.

Paul Iusztin's avatar

yes! if latency is not a problem, this solution works amazingly for me. What are you working on?

Alex Openstone's avatar

https://labintrace.com/tour I am building an AI governance architecture for regulated environments using publicly available cited CAP/CLSI/Hl7 standards. This framework also incorporate a Qdrant RAG pipeline for locality based lab SOPs, evidence based reg reviews which are encoded into a knowledge based structure.

Tap's avatar

I think that the expensive tasks (such as ingest data, embedding a large of data, gen metadata) will not good to use harnesses agent.

Paul Iusztin's avatar

They still have their place, for example, using RLM + progressive disclosure works, but it's super slow. So if you want something snappy or cheaper, you still need to implement a normal index on top of it.

But for knowledge work, using RLMs is a gold mine as you can query any type of data with minimal effort