Lexii.ai was a precursor to more well known search assistants today such as Bard and Perplexity. Our system translated user queries into Google searches, returned scraped results, and synthesized them into answers while supporting multi-turn dialog.
Responding to the chat-with-your-data craze, Lexii pivoted to an embeddable chat widget, delivering answers to questions based on an easy to manage knowledge base. We built a RAG system from the ground up, with vector embeddings for each knowledge base item and chain of thought prompting to deliver the best answer.
Lexii's final pivot turned profitable as an article generator based on GPT-4. Although it appears straightforward, the implementation was very complex. It solved a number of painful problems for users — specifying an exact word count, avoiding plagiarism, avoiding the "hollow" sound of ChatGPT, formatting, and more. Our solution involves RAG (with web scraping), agents for writing and editing, and complex business logic, driving chain of thought prompts.