CVision

CVision is an AI-powered CV and career analysis platform. Users upload a resume, paste a job description, or ask career questions through a chat interface — and receive real-time streaming feedback. No sign-up required.

Live Demo

• Landing page: https://cvision-llpy.onrender.com
• Chat interface: https://cvision-llpy.onrender.com/chat.html

Features

• AI-Powered CV Analysis — Upload your resume and a job description to get a detailed match score, skill gap analysis, and actionable recommendations
• Real-Time Chat — Stream responses token by token over WebSocket for a fluid conversational experience
• Document Parsing — Supports PDF and DOCX uploads with automatic text extraction
• No Sign-Up Required — Start analyzing immediately with zero friction
• Privacy-First — All documents processed securely; no account or data retention required
• ATS Keyword Detection — Identifies missing keywords that could hurt your application in applicant tracking systems

Key Components

Orchestrator Agent Pipeline

The system routes messages through an agent pipeline coordinated by server/routes/chat_router.py:

• Text messages → agent_handler.py → CVisionAgent → Flowise AI (SSE streaming)
• File uploads → Celery cv_analyzer task → n8n Parser API (PDF/DOCX parsing) → CVisionAgent → Flowise AI

The CVisionAgent (server/agent/cvision_agent.py) wraps external AI services — Flowise for LLM-powered analysis and n8n for document parsing — streaming results back chunk by chunk.

Redis Stream (Message Buffer)

Redis Streams decouple the producer (agent/Celery) from the consumer (WebSocket):

• Producers — agent_handler.py writes each SSE chunk from Flowise to task_{task_id} via XADD; the Celery task writes status updates and parsed results
• Consumer — the WebSocket handler blocks on XREAD polling, forwarding each event to the frontend
• Stream pattern — task_{task_id} per session; event fields: {event, data}

Uses Upstash (serverless Redis) with sync Redis for writing and async Redis for the polling loop.

WebSocket (Real-Time Streaming)

• Endpoint: ws://host:port/ws/v1/chat/
• Protocol: Bidirectional JSON messages
• Flow: Client sends text/binary → server writes events to Redis Stream → server polls and forwards to client → client renders tokens incrementally
• Frontend — ChatClient class in frontend/chat-core.js manages auto-reconnect, stream bubble accumulation, and file-to-base64 conversion

Tech Stack

• Frontend: HTML, CSS, JavaScript (vanilla)
• Backend: FastAPI (Python)
• WebSocket: Real-time bidirectional streaming
• Message broker: Redis Streams (Upstash)
• Background workers: Celery (Redis broker)
• External AI: Flowise (LLM), n8n (document parsing)