Skip to main content
Apolla AI
OperationsFeatured

Enterprise document search

Unified semantic search across policies, product and project documents, with answers that cite the source text and permissions that follow the user.

Sample input and output

Sample input

Under the travel accommodation policy, what is the daily cap for director level in tier-1 cities? Which document says so?

Sample output

The accommodation cap for director level in Beijing, Shanghai, Guangzhou and Shenzhen is RMB 800 per night (tax included). Source: Travel Expense Policy (2025 revision), clause 4.2, page 6 of the attachment.

Details

Company documents are scattered across the OA system, shared drives, wikis and email attachments; an employee typically switches between 3–4 systems to find the original text of one policy. This template connects those sources to a single knowledge base so people can ask in natural language and get answers with sources.

Retrieval combines vector and BM25 recall with a reranker. Answers are grounded strictly in the retrieved passages and cite where each one came from. Users can only retrieve documents they are allowed to see: out-of-scope content is filtered at retrieval time, not masked after generation.

A typical deployment takes 1–2 weeks: connect document sources, map permissions, and run a 100–300 item evaluation set to calibrate recall.

  • Parses and incrementally syncs Word, PDF, PowerPoint, Excel, web pages and OA attachments
  • Answers include the source passage and a document link for one-click verification
  • Retrieval-level permission filtering by org structure and document ACLs
  • Built-in recall and faithfulness evaluation that keeps scoring after go-live

How close is this template to your use case?

Most customers go live after changing 20–40% of a template. Book a workshop and we will assess it together.