Skip to main content

ARIA: Autonomous
Research Assistant

A 5-stage stateful agentic RAG pipeline built with LangGraph. ARIA goes beyond simple single-turn vector retrievals by planning query structures, executing concurrent multi-source scrapers, self-critiquing generated content, and compiling comprehensive PDF reports.

Live Demo / PWA Windows 3.20MB APK 2.27MB Source Code
Core Infrastructure
LangGraph Orchestration & State Management
ChromaDB Local Vector Storage & Memory
asyncio Concurrent Gathering
ReportLab Automated PDF Generation
Flagship System

ARIA

Autonomous Research & Intelligence Assistant. A 5-stage agentic RAG pipeline built with LangGraph, reducing retrieval latency by 63% via concurrent hybrid search across Wikipedia, OpenAlex, arXiv, and DuckDuckGo.

GitHub Repo
ARIA Dashboard // React SPA // JWT Auth
Session Log
> Executing query...
> 4 sources active
> Latency: 4s (~63% reduction)
LangGraph Orchestrator Pipeline
Plan
Retrieve
Synthesize
Audit
Publish
63% Latency Reduction

Concurrent hybrid retrieval (`asyncio`, `aiohttp`) cutting retrieval time from ~11s to ~4s across 4 knowledge sources.

11/11 Unit Tests Passing

Ragas-based eval suite verifying faithfulness, answer relevancy, and context precision. 50% leakage tests passed.

Full Cross-Platform Stack

Azure VM (Ubuntu + Caddy). React/Tailwind SPA, PyWebView Desktop app, Gradle/TWA Android APK. PDF exports via PyMuPDF.