Query Now - MultiDoc RAG

An extensible, production-grade platform that unifies knowledge from multiple sources into an intelligent retrieval system. Transform unstructured documents into queryable knowledge graphs with autonomous AI agents that dynamically orchestrate vector search, graph traversal, and logical filtering for optimal information retrieval.

---

Continuous Monitoring For Safe Commits: This project is monitored and tested by Oggy for code quality and safe commits built by me.

Architecture

See detailed architecture →

1. Document-to-Graph Pipeline

Document Upload (PDF/DOCX/TXT)
    ↓
Text Extraction & Chunking
    ↓
LLM-Powered Ontology Generation
    ↓
OpenAI Embeddings
    ↓
Parallel Processing:
    ├─→ Pinecone Vector Store (semantic search)
    ├─→ Neo4j Knowledge Graph (entity relationships)
    └─→ Redis Cache (query optimization)

2. Agentic Retrieval System

Natural Language Query
    ↓
Query Analysis Agent
    ↓
Dynamic Strategy Selection:
    ├─→ Vector Similarity Search (OpenAI embeddings)
    ├─→ Graph Traversal (Neo4j Cypher)
    └─→ Logical Filtering (metadata/attributes)
    ↓
Multi-Step Reasoning & Iterative Refinement
    ↓
Response Graph Generation
    ↓
Streaming Response with Reasoning Chain

3. Core Components

Embedding Layer

• Model: OpenAI text-embedding-3-large
• Dimensions: 3072
• Usage: Documents, entities, relationships, queries

Vector Store (Pinecone)

• Semantic similarity search
• Hybrid search capabilities
• Metadata filtering

Graph Database (Neo4j)

• Entity resolution & deduplication
• Relationship extraction
• Ontology management
• Cypher query generation

Cache Layer (Redis)

• Query result caching
• Session management
• Performance optimization

AI Orchestration

• GPT-4 for ontology generation
• GPT-4 for entity extraction
• Autonomous agent routing
• Multi-tool reasoning

4. Key Features

Automatic Ontology Generation - LLM extracts entities, relationships, hierarchies
Entity Resolution & Deduplication - Intelligent merging of similar entities
OpenAI Embeddings - 3072-dimensional vectors for all graph elements
Agentic Retrieval - Dynamic tool selection across vector/graph/filter methods
Visual Knowledge Graphs - Interactive graph visualization
Streaming Responses - Real-time reasoning chains
Multi-Step Reasoning - Iterative query refinement

---

Tech Stack

Frontend: Next.js 15, TypeScript, Tailwind CSS
Auth: Kinde
Storage: Appwrite
AI: OpenAI GPT-4 (embeddings + generation)
Vector DB: Pinecone
Graph DB: Neo4j
Cache: Redis

---

Quick Start

git clone <repo-url>
cd query-now
pnpm install
pnpm run dev

Configure .env.local with API keys for OpenAI, Pinecone, Neo4j, Redis, Appwrite, and Kinde.