aisearchnatural languagesemantic search

How Search Works

2 min read

Archevi uses advanced AI technology to understand your questions and find relevant information across all your documents. Unlike traditional keyword search, our AI understands meaning and context, making it feel like you're asking a knowledgeable assistant rather than searching a database.

How It Works

When you ask a question, Archevi's AI processes it through several stages:

  • Understanding -- The AI analyzes your question to understand what you're really asking, even if you phrase it casually
  • Understanding Your Questions -- Rather than matching keywords, the AI searches for documents with similar meaning and relevance
  • Checking the Source -- The AI considers the context of your documents to provide the most relevant results
  • Finding the Answer -- When possible, the AI extracts and highlights the specific answer from your documents

Natural Language vs Keywords

The key difference between AI search and traditional search is that you can ask questions naturally:

Traditional Keyword Search

  • You search: "passport expiry 2025"
  • Requires exact word matches
  • Returns documents containing those words

search

  • You ask: "When does my passport expire?"
  • Understands you want expiry date information
  • Finds relevant documents even if they don't contain the word "expire"
  • Extracts the actual date from your passport document

What Makes Our AI Different

Archevi's AI is specifically designed to understand family documents. It recognizes:

  • Document types (passports, insurance cards, contracts, receipts)
  • Important dates (expiry dates, renewal dates, effective dates)
  • Key entities (names, addresses, policy numbers, amounts)
  • Relationships (connecting related documents together)

Privacy and AI Processing

Your privacy is paramount. Before any query reaches an AI model, Archevi's boundary anonymization system (powered by Microsoft Presidio) automatically replaces personal information with realistic surrogates. Names become fake names, emails become fake emails, and sensitive identifiers are blocked entirely.

The anonymized queries are processed by Groq (using Llama language models) for generating answers and Cohere for semantic search and document retrieval. Both providers have contractual no-training commitments -- they never use your data to train their models. And because of boundary anonymization, they only ever see surrogates, never your real personal information.

All your documents are stored on Canadian servers (DigitalOcean Toronto) and never leave the country. Only anonymized query text crosses the border to AI providers. Learn more in our data protection guide and on our security page.

Getting the Best Results

For tips on writing effective search queries, see our search tips guide. To understand the different types of questions you can ask, check out types of questions.