Asking Your Organization's Documents in Persian: What Is Searchable and How Sources Are Shown

An organization's knowledge actually lives scattered across folders, Excel files, scans of contracts, meeting recordings and phone calls. When someone asks "what were the warranty terms in that contract?" or "what did we decide in last week's meeting?", finding the answer usually depends on one person's memory or hours of searching. An intelligent knowledge base aims to make all of that queryable in Persian.

What you can ask

AfranDanesh (Afrandeh's AI) brings five sources under one roof:

• Documents: PDF and TXT files.
• Excel: XLSX and XLS files.
• Photos and scans: PNG, JPG, BMP and TIFF, with text read from the image (OCR).
• Video and audio: formats such as MP4, MOV, MKV, AVI, MP3, WAV, M4A and OGG, with automatic speech transcription.
• Phone-system calls: recorded calls from Asterisk, FreePBX and Issabel phone systems.

How a question reaches an answer

Each question goes through four stages: semantic search (finding material with the same meaning, even in different words), keyword search, result fusion and final ranking. The result shows the most relevant passages with their exact source and relevance, so you can follow any answer back to the document or the moment in the call.

Why "built for Persian" matters

General-purpose tools often struggle with Persian details. In AfranDanesh half-spaces, Arabic "ی" and "ک" and the three digit systems (Persian, Arabic and Latin) are treated as one; a phone call's ID card is built from the recorded file's name; and homophone letters are corrected for more accurate search.

Where the data stays

Every component, from the database and search to the speech models, runs on your organization's own server; documents, transcripts and question history never leave your infrastructure. Each company has its own knowledge base, each user searches only the companies they are allowed, and there are three roles: system admin, company admin and regular user.

One honest note

Today, instead of writing a full answer like a generative chat, the system shows the most relevant passages with their source. A local large language model that writes full answers is in progress, and it too will run without any external service.

If you want it connected to your systems

If your goal goes beyond documents and you want phone-system, finance, inventory and office-automation data connected to a dedicated RAG, see Afrandeh Enterprise RAG. To see AfranDanesh on your own samples, request a demo.