ai-agent-smart-assist
Basic Information
This repository is a personal LangChain-powered AI agent designed to let a user upload documents, classify text, build a searchable knowledge base, and query uploaded content. It focuses on simple, user-facing interactions: switchable modes for classifying incoming text or ingesting files, automatic routing to the appropriate processing logic, and question answering over ingested documents. The system chunks uploaded files, indexes embeddings in a FAISS vector store, and uses sentence-transformer embeddings with a hosted model for retrieval-augmented responses. The project includes both a backend API and a frontend so a nontechnical user can run ingestion and query flows without writing code.
Links
Stars
209
Language
App Details
Features
The project implements automatic text classification and tool routing, a document ingestion pipeline that chunks and parses files using Unstructured, and a FAISS-based vector store for efficient retrieval. It supports multi-file uploads, a toggleable UI for Classify versus Ingest modes, pipeline execution, and viewing or exporting full results. The stack includes LangChain tools, Sentence Transformers (MiniLM) for embeddings, FastAPI for the backend, and a Next.js plus Tailwind frontend. The README shows run commands for the backend using uvicorn and for the frontend via npm run dev. The repo also mentions using the Gemini model from Vertex AI and plans for adding RAG-style chat and namespace support.
Use Cases
This agent helps users turn scattered documents into a searchable, question-answerable knowledge base so they can quickly get precise answers from their own files. Classification automates routing to the right processing logic so different document types trigger appropriate actions. Ingest mode handles chunking and embedding storage, enabling retrieval-augmented generation for context-aware answers. The combined backend and frontend lower the barrier to use by allowing uploads, pipeline runs, and exports through a web UI. Built-in technologies like FAISS and MiniLM make searches fast and relevant. Roadmap items such as chat-style RAG flows and multi-project namespaces suggest it can grow into a more conversational and multi-tenant knowledge assistant.
-
$$$$ -
$$$$