跳到正文
DEV Community· Saad Wasim·· 15 小時前AI 評分28

我打造了 Deskwise:供公司文件使用的私有 AI 助理(Django + Next.js)

I built a private AI assistant for company documents (Django + Next.js)

AI 導讀

Deskwise 是作者打造的私有 AI 助理,能根據公司文件回答問題並標示來源頁面。系統按公司、部門或指定人士設定存取權限,並記錄上載、提問及下載;搜尋結合 70% 向量相似度與 30% BM25。產品已可運作,但尚未有收入或客戶;作者正出售完整原始碼及知識產權,並提供交接。

正文

Cover image for I built a private AI assistant for company documents (Django + Next.js)

Saad Wasim


I built Deskwise, a private AI assistant that answers questions from a company's own documents.

What it does

Upload handbooks, policies and contracts (PDF, Word, Excel, scans).
Ask in plain language, in English or Roman Urdu. You get the answer plus the source page.
Employees only get answers from documents they're allowed to see. Access is set per company, department, or named person.
Every upload, question and download goes in an audit log, which can be exported.
It can fill in leave requests and HR letters and export them as Word or PDF.
How it works Documents are split into chunks and indexed. Search blends vector similarity (70%) with BM25 keyword matching (30%). Permissions are enforced in the query layer, not just in the UI. The answer is generated by the AI provider you configure (Ollama, OpenAI, Gemini or AWS Bedrock) and cites its sources.

Stack: Django, Next.js, Postgres with pgvector (SQLite for local use), Redis, Docker Compose.

Status: it's a working product, but pre-revenue with no customers. I'm a student and I'm selling the full source code and IP, with a handover.

Message me for a demo login
Link: https://deskwise.saadops.site
Walkthrough: https://youtu.be/qNLAX96X5j0

Happy to answer questions about the design, especially how the permissions work with retrieval.

來源:DEV Community · dev.to