Q&A Bots Enterprise SDK

The Q&A Bot SDK is designed for precise, single-turn question-answering. It is highly optimized for integration into customer support portals, helpdesk FAQ sections, and lookup applications such as employee handbooks, compliance policies, and product documentation.

Unlike the full conversational ChatBot, the Q&A Bot does not track long-term memory or message history. This stateless behavior reduces latency, lowers API costs, and ensures clean response isolation for standalone lookup queries.

Getting Started

Initialize the Q&A Bot with a knowledge database or folder of documents.

Simple Q&A Example

import { QABot } from '@ai-horizon/sdk';
 
const qaBot = QABot.initialize({
  input_dir: "./compliance_policies",
  system_prompt: "You are a strict compliance checking assistant. Cite policies exactly."
});
 
// Perform a single-shot query
const result = await qaBot.query("What is the policy for business travel reimbursement?");
console.log("Answer:", result.answer);

Supported Source Formats

Just like the ChatBot, the Q&A Bot supports multiple target document formats:

// Init from PDF
const qaFromPdf = QABot.pdf_qa({
  input_files: ["./docs/employee_handbook.pdf"]
});
 
// Init from a Webpage
const qaFromWeb = QABot.webpage_qa({
  urls: ["https://support.mycompany.com/faq"]
});
 
// Init from a Word doc
const qaFromDocx = QABot.docx_qa({
  input_files: ["./specs/product_user_guide.docx"]
});

Key Configurations for Q&A Bot

The initialization methods support configuration options tailored for high-accuracy lookups:

Parameter	Type	Default	Description
`input_dir`	`string`	`null`	Folder path containing reference documents.
`input_files`	`Array`	`[]`	List of files to index.
`similarity_top_k`	`number`	`3`	Number of relevant chunks retrieved to answer the question.
`score_threshold`	`number`	`0.7`	Minimum similarity score required to include a document source. Helps reduce hallucinations.
`system_prompt`	`string`	`null`	Guides the system's focus and answering rules.
`stream`	`boolean`	`false`	When true, response chunks are streamed for low-latency client rendering.

Handling "I don't know" Responses

One of the most important aspects of enterprise compliance is preventing the AI from hallucinating. You can instruct the Q&A Bot to only reply based on retrieved sources:

const qaBot = QABot.pdf_qa({
  input_files: ["./docs/security_audit.pdf"],
  system_prompt: "Answer the question ONLY based on the provided context. If the answer is not in the context, reply 'Information not found in database.'"
});
 
const result = await qaBot.query("What was our 2024 revenue?");
// If 2024 revenue is not in the audit PDF, it will return "Information not found in database."

ChatBots Data Analysis Bot