AI-Powered Workflow Agent in a Sandboxed Environment

Project Difficulty 🚀

Intermediate

Target Audience 🎯

Computer Science Students

Electrical and Computer Engineering Students

Description 📌

This project focuses on developing an AI-powered agent that automates repetitive and complex workflow tasks in a specific domain, such as software development, project management, or cloud operations. The AI agent will be deployed in a sandboxed environment to ensure safety, prevent unintended consequences, and provide controlled automation with strict oversight.

Key Objectives 🔍

Task Identification & Domain Selection: Choose a workflow domain (e.g., DevOps, educational content moderation, research assistance) that benefits from automation while avoiding sensitive applications like finance.

Data Collection & Preprocessing: Gather workflow logs, user interactions, and process documentation in a controlled manner to ensure safety and privacy.

Sandboxed Execution: Deploy the AI agent in a secure, isolated environment to test and validate automation before live system integration.

Integration with Existing Tools: Connect the AI agent with popular workflow tools like JIRA, Slack, Jenkins via APIs.

User Interface & Interaction: Develop a chatbot or dashboard to enable manual overrides, status tracking, and recommendation validation while enforcing safety checks.

Performance & Safety Evaluation: Measure automation effectiveness using KPIs such as time saved, accuracy, efficiency, and adherence to safety policies.

Expected Outcomes 🎯

✅ An AI-driven agent automating repetitive tasks in a sandboxed environment. ✅ A performance & safety evaluation report showcasing efficiency improvements, cost reduction, and risk mitigation. ✅ A deployable prototype demonstrating real-world application and scalability.

Estimated Project Duration ⏳

0-2 months

Team size: 2-5 participants

Hardware / Software Requirements 🛠

The tools below are simply suggestions, you are free to use the tools of your choice.

🔹 Programming Languages

Python

JavaScript (for front-end, if needed)

🔹 APIs & Automation Tools

FastAPI, Flask, Selenium

🔹 Database

PostgreSQL, MongoDB

🔹 Cloud Services

Arm-based servers on AWS, Google Cloud, or Azure

🔹 Sandboxing & Security

Docker-based isolation, SELinux, Kubernetes namespaces

🔹 Hardware

Cloud compute resources (Arm-based AWS, GCP, or Azure)

API access to target workflow tools

Previous Submissions

  1. AI to Solve Maths Example Sheets at University of Cambridge. (Finley Stirk, Eliyahu Gluschove-Koppel and Ronak De) - https://github.com/egkoppel/example-papers

  2. AI that interprets user requests, generates circuit descriptions, creates LTSpice ASC code, and iteratively refines circuit designs using a combination of GPT-based language models, a vision analysis module, and LTSpice simulation. (Gijeong Lee, Bill Leoutsakos) - https://github.com/BillLeoutsakosvl346/ElectroNinjaRefined

Resources 📚

Arm Learning paths

Benefits / Prizes 🏆

🎓 Hands-on experience in AI-driven automation with a strong focus on sandboxed execution & safety.

🚀 Exposure to API integration & cloud services, highly relevant for industry roles in AI, automation, and cloud computing.

📃 Standout projects may receive internal referrals for relevant positions at Arm!

🏅 Recognized badge for approved submissions, perfect for listing on your CV or LinkedIn!