Solution

Quantiphi proposed baioniq, our in-house generative AI platform for legal documents, capable of reading, ingesting, summarizing, and redlining contracts.

The project handled four types of contracts: Letter of Indemnification, Academic, Contract Research Organization, and Non-Contract Research Organization.

We developed a three-layer ML solution for redlining:

Clause Extraction: Finetuned text-bison model to extract relevant clauses from contracts.
Retriever: Built a retriever layer to extract the right clauses based on rules provided by the UH team.
Redlining: Finetuned text-bison model to perform contract redlining.

Results

Quantiphi implemented a standard train, test, and evaluation strategy for our model, which is currently being tested by the customer on the UI. Preliminary results show 60-65% accuracy for the redlining layer, while the clause extraction and retriever layers achieve over 85% accuracy.