Autonomous AI Knowledge Extraction
TURN
TEXT INTO
INTELLIGENCE.
AURA is an autonomous AI agent that reads any unstructured text and extracts entities, relationships, and semantic connections — building a persistent knowledge graph in real time.
Powered by Llama 3.1 via LangGraph, stored in CockroachDB with pgvector, and rendered as an interactive physics-driven 3D force graph directly in the browser.
01/
LLM MODEL
LLAMA 3.1
02/
DATABASE
COCKROACHDB
03/
BACKEND
FASTAPI
04/
DEPLOY
AWS EC2
HOW AURA
WORKS.
Step 01
INPUT
You paste any unstructured text into the chat — an article, raw notes, a Wikipedia paragraph, or any concept you want to map. AURA's minimal interface accepts raw human language with no formatting requirements.
Step 02
EXTRACT
AURA's LangGraph-based agentic pipeline sends your text to Llama 3.1 (hosted on Groq). The LLM autonomously identifies named entities (people, concepts, technologies, events) and the relationships between them. No manual tagging or schema definition is needed.
Step 03
RENDER
Extracted nodes and edges are persisted into CockroachDB (a globally distributed SQL database). The frontend then fetches this graph data and renders it as an interactive, physics-driven 3D force-directed graph using react-force-graph-3d — right in your browser.
THE
MECHANISM.
Step 01
USER INPUT → NEXT.JS FRONTEND
The user types unstructured text into the chat interface built with Next.js 15 (App Router). On submit, the frontend sends a POST request to /api/chat — a Next.js API route that acts as a proxy.
POST /api/chat → { text: "Albert Einstein developed the theory of relativity..." }Step 02
NEXT.JS PROXY → FASTAPI BACKEND
The Next.js API route forwards the request to the FastAPI backend server hosted on AWS EC2. This proxy layer exists to avoid exposing the backend IP directly to the browser and to handle CORS.
Next.js (Vercel) → HTTP POST → FastAPI (EC2: http://<ip>:8000/chat)Step 03
FASTAPI → LANGGRAPH AGENT
FastAPI receives the text and passes it to a LangGraph-based agentic pipeline. LangGraph orchestrates the workflow — it manages state, decides which tools to call, and handles retries. The agent has access to tools for entity extraction and database writes.
agent.invoke({ input: userText }) → LangGraph state machine runsStep 04
LANGGRAPH → LLAMA 3.1 (GROQ)
The LangGraph agent sends the user's text to Llama 3.1-70B hosted on Groq's ultra-fast inference API. The LLM is prompted to extract structured data: a list of entities (nodes) with types and a list of relationships (edges) with labels connecting them.
Groq API → Llama 3.1-70B → Returns JSON: { nodes: [...], edges: [...] }Step 05
EXTRACTED DATA → COCKROACHDB
The extracted nodes and edges are written into CockroachDB — a distributed, cloud-native SQL database. Nodes are stored in a 'nodes' table with ID, label, and group fields. Edges are stored in an 'edges' table with source, target, and label. Duplicate detection prevents redundant entries.
INSERT INTO nodes (id, label, group) ... INSERT INTO edges (source, target, label) ...Step 06
RESPONSE → FRONTEND → 3D GRAPH
FastAPI returns the agent's natural language response along with counts of extracted entities and relationships. The frontend then fetches the full graph via GET /api/graph, which queries CockroachDB for all nodes and edges. This data is fed into react-force-graph-3d, which renders a live, interactive, physics-simulated 3D knowledge graph.
GET /api/graph → { nodes: [...], links: [...] } → Three.js 3D RenderHOW IT WAS
BUILT.
NEXT.JS 15
Frontend
App Router, Tailwind v4, react-force-graph-3d
Vercel Edge
FASTAPI
Backend
Python async, LangGraph orchestration, REST APIs
AWS EC2
LLAMA 3.1
LLM Brain
70B model, structured entity extraction
Groq API
COCKROACHDB
Database
Distributed SQL, stores nodes/edges, pgvector
Cloud SQL
LANGGRAPH
Agent
State machine for autonomous workflows
Framework
GROQ API
Inference
Ultra-fast low latency LLM inference
LPU
FRAMER MOTION
Animation
Declarative brutalist UI animations
Library
THREE.JS
3D Engine
Hardware-accelerated graph rendering
WebGL
KEY
FEATURES.
AUTONOMOUS EXTRACTION
No manual tagging required. The AI agent autonomously decides what entities to extract and how they relate to each other.
3D KNOWLEDGE GRAPH
Interactive, physics-driven 3D force-directed graph rendered in the browser. Zoom, pan, rotate — explore your knowledge visually.
PERSISTENT STORAGE
All extracted knowledge is permanently stored in CockroachDB. Your graph survives page refreshes and server restarts.
REAL-TIME UPDATES
Graph updates live after every extraction. New nodes and edges appear instantly without requiring a page reload.
INCREMENTAL BUILDING
Each new input adds to the existing graph. Over time, your knowledge graph grows richer and more interconnected.
ENTITY MANAGEMENT
View, search, and delete individual nodes and edges. Full CRUD operations on your knowledge graph through the Database Context panel.
YOUR GRAPH
STARTS WITH
ONE SENTENCE.
No setup. No configuration. No schema definition. Type one sentence and watch AURA construct a living knowledge network.