lyxhnu

lyxhnu / Skill-First-Hybrid-RAG

Public

一个可运行的 `Skill-first + Vector-augmented + LangGraph` RAG 系统，支持多模型厂商、分层记忆和 Web 聊天界面。

23
1
100% credibility
Found Mar 10, 2026 at 18 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

A local AI chat system that searches documents using specialized skills for routing, hybrid dense-sparse retrieval, structured analysis for spreadsheets, and multi-turn memory.

How It Works

1
🔍 Discover Smart Document Helper

You find this friendly tool that helps search and answer questions from your own documents like PDFs and spreadsheets.

2
🗂️ Gather Your Files

Put your documents into a simple folder so the helper knows where to look for answers.

3
🔗 Link Thinking Partners

Connect free smart services that help the tool understand and reason about your questions.

4
🚀 Wake Up the Chat

With one easy start, your personal document assistant comes alive on a web page ready to chat.

5
📚 Feed It Your Knowledge

Tell the assistant to learn from your files, and it builds a smart map of everything inside.

6
💬 Ask Away Naturally

Type questions like 'Who are the top shareholders?' and get clear answers with exact sources.

✅ Smart Answers Forever

Enjoy reliable replies every time, improving as you chat, with proof from your own files.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 18 to 23 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is Skill-First-Hybrid-RAG?

This Python-based RAG system prioritizes skill-first retrieval—routing queries to specialized skills for PDFs, Excel analysis, and JSON FAQs—before falling back to vector-augmented hybrid search with FAISS dense and BM25 sparse. Orchestrated via LangGraph, it handles local knowledge bases with multi-model support (OpenAI-compatible APIs), layered conversation memory, and strict no-evidence refusal with citation checks. Users get a web chat UI, REST API for /query and /ingest, plus an e-commerce loop that captures unanswered queries for human FAQ updates.

Why is it gaining traction?

It stands out by enforcing skill-first precision over naive vector RAG, delivering grounded answers from structured docs like Excel sheets or PDF pages without hallucination risks. The hybrid mode blends skills with vectors only when needed, plus built-in memory for multi-turn chats and automatic feedback for gaps. Developers appreciate the quick-start with env vars for models like GLM or Qwen, and the web interface for testing without coding.

Who should use this?

AI engineers prototyping enterprise RAG for financial reports or inventory data in mixed PDF/Excel/JSON corp knowledge bases. E-commerce teams building customer service bots that learn from misses via FAQ ingestion. Python devs experimenting with LangGraph workflows needing production-ready retrieval with web demo.

Verdict

Solid docs and runnable out-of-box make it worth forking for custom RAG, despite 19 stars and 1.0% credibility signaling early maturity—expect tweaks for scale. Try if skill-first hybrid fits your docs; skip for plug-and-play alternatives.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.