Cogni-Flow

.Cogni-Flow: The Autonomous AI Apprentice

Submission for the Galuxium Nexus V1 Hackathon Focus Domains: Autonomous AI Agents, Developer Productivity Tools Demo Video: [LINK TO YOUR 2-MINUTE VIDEO HERE] Presentation Deck: [LINK TO YOUR GOOGLE SLIDES HERE]

The Core Challenge

Expert knowledge is "trapped." The most valuable, complex workflows (in coding, design, and operations) are stuck in the heads of senior-level experts. They are difficult to document and impossible to automate with simple scripts because they rely on context and intuition.

Our Solution: Cogni-Flow

Cogni-Flow is an "AI apprentice" agent that learns to perform complex tasks simply by observation.

It's a simple 3-step process:

WATCH: The user clicks "Start Recording." The agent observes their screen, clicks, and keyboard actions as they perform a complex task.

LEARN: The user clicks "Stop & Learn." Cogni-Flow sends the action log to a Generative AI model (Google's Gemini 2.5) to analyze the logic and intent behind the actions. The AI then writes a new, autonomous Python script.

AUTOMATE: The user can now click "Run Workflow," and the agent will execute that newly generated script, performing the entire complex task on its own.

This redefines human-computer interaction by using passive observation as an input and accelerates productivity by automating the "un-automatable."

The Working Prototype

This repository contains the working prototype, app.py. It's a desktop agent built in Python that fully demonstrates the "Watch, Learn, Automate" loop.

Tech Stack Core Logic: Python Desktop UI: customtkinter AI "Brain": Google Gemini (google-generativeai) Sensing (Recording): pynput (for mouse/keyboard) & mss (for screen) Acting (Automation): pyautogui

Built With

• customtkinter
• openai
• pyautogui
• python