Autonomous Vision-Enabled Agentic Platform for Web Research & Competitive Intelligence
Decompose complex queries into parallel browser agents. Get real answers, not hallucinations.
ARIA is not a chatbot. It's an autonomous multi-agent system that browses the real web in parallel to answer complex research and competitive intelligence queries.
When you give ARIA a goal, it:
- Decomposes your query into 1–6 parallel sub-tasks using Groq LLaMA 3.3
- Spawns real browser agents — each navigating live websites simultaneously
- Extracts structured data from pricing pages, job boards, news sites, financial dashboards
- Synthesises all results via Gemini 1.5 Pro into a coherent report
- Saves everything to PostgreSQL — full session history, resume profiles, watchlists
The key differentiator: real browsers, real pages, real-time data — no hallucination, no stale training data.
| Mode | What It Does |
|---|---|
| 🔍 Research Analyst | Multi-source deep research with source citations |
| 💼 Career Copilot | Resume parsing + live job search + cover letter generation |
| 💰 Money Agent | Real-time stock prices, analyst ratings, ETF comparisons |
| ⚡ Form Executor | Autonomously fills and submits real web forms |
| 🤖 Assistant | General-purpose browser-powered AI assistant |
| 📝 Content Agent | Research-backed content generation with live sources |
| 🎯 Job Autopilot | Discovers and applies to jobs automatically |
| 👁️ Competitor Watch | Monitors competitor pricing and product changes |
- Decentralized multi-agent design — agents run in parallel, not sequence
- Each agent has its own browser session, task scope, and result stream
- SSE-based real-time streaming — watch agents work live
- Multimodal AI detects semantic changes in websites via screenshot comparison
- Bounding box localization of changed elements
- Used for competitor monitoring and content change detection
- Autonomous career agent that parses resumes into structured profiles
- Performs intelligent job discovery and form automation
- Supports 11 Indian languages via Sarvam AI (voice input + TTS output)
- Real-time browser automation with stealth profiles
- Reliable navigation across dynamic and JS-heavy sites
- Handles anti-bot protection via profile rotation
| Layer | Technology |
|---|---|
| Frontend | Next.js 15, TypeScript, Tailwind CSS |
| LLM (Planning) | Groq LLaMA 3.3 70B (ultra-fast inference) |
| LLM (Vision/OCR) | Gemini 1.5 Pro |
| Browser Automation | TinyFish real browser agents + SSE streaming |
| Voice | Sarvam AI (11 Indian languages, TTS + STT) |
| Database | Supabase PostgreSQL |
| Auth | Supabase Auth |
User Goal (text or voice)
│
▼
Groq LLaMA 3.3 (Intent Parsing + Task Decomposition)
│
▼
┌─────────────────────────────────┐
│ Agent Swarm (parallel) │
│ Agent 1 │ Agent 2 │ Agent 3 │
│ Browser │ Browser │ Browser │
└─────────────────────────────────┘
│
▼
Gemini 1.5 Pro (Synthesis + Vision)
│
▼
Structured Report + Voice Output (Sarvam AI)
│
▼
Supabase PostgreSQL (Session History)
git clone https://github.com/pritpatel2412/ARIA
cd ARIA
pnpm install
cp .env.example .env # Add your API keys
pnpm devRequired env vars: GROQ_API_KEY, GEMINI_API_KEY, SUPABASE_URL, SUPABASE_SERVICE_ROLE_KEY, SARVAM_API_KEY
Prit Patel — B.Tech CSE @ CHARUSAT University GitHub · LinkedIn