Build a Document Assistant in Seconds
- Ultra-fast Responses
- Upload your documents
- Get answers with verified citations
- Query on-screen or via API
Who/What is this for?
- Build an AI Assistant REALLY quickly.
- You want to save time setting up RAG/embeddings/retrieval/databases.
- You want to quickly prototype an AI app that includes document retrieval.
- You are building a chatbot and need an LLM endpoint that is grounded in a set of your documents.
What do you get?
- An API endpoint for your documents in just a few clicks.
- Responses that include verified citations.
What’s in it for you?
- Save hours of developer time.
- Mitigate hallucinations from language model responses.
- Very fast responses.
How does it work?
- Documents are cleaned and chunked.
- Retrieval is done for each query using both BM25 (advanced key-word search) and cosine-similarity on document chunks.
- Relevant chunks are injected into the prompt template.
- A large language model is prompted to return an answer with citations.
- Citations are verified against the underlying documents.
- A response is provided, including citations.