Utilize Cleanlab's trustworthy language model (TLM) to catch LLM hallucinations

[AshishSardana]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

Cleanlab provides trustworthy language model (TLM) that adds a trustworthiness score to every LLM response, letting one know which outputs are reliable and which ones need extra scrutiny.

An integration of Cleanlab's TLM with LangChain would allow developers to take advantage of the trustworthiness in agentic/RAG applications.

Motivation

As with most LLM based applications, the most popular Enterprise use-cases are based on RAG. The trustworthiness_score is useful in enhancing reliability by - routing requests appropriately, tracking unchecked hallucinations, and more.

Cleanlab's TLM has gained traction proving its usefulness.

Proposal (If applicable)

Raised a working basic PR #26412 that integrates Cleanlab in langchain-community package, with an example notebook.