Or you can build something using a combination of LLM models (OpenAI, Cohere, Hugging Face) and Pinecone (offers keyword-aware semantic search which is what you probably want for domain-specific content like legal… I am affiliated).
I trained my own models, actually. Also tried many of the vector search engines out there. But ended up with a custom solution using hnswlib and SQLite to get subsecond speed with millions of documents.
Or you can build something using a combination of LLM models (OpenAI, Cohere, Hugging Face) and Pinecone (offers keyword-aware semantic search which is what you probably want for domain-specific content like legal… I am affiliated).