AWS Builder Center

Bridging the Cloud to the Edge: Building BotOps at the AWS Hackathon

How my event of First Commit AWS has helped me discover

ML Engineer/ ML Ops
The energy at a hackathon is always a unique mix of adrenaline, frantic debugging, and the thrill of solving impossible problems. For this AWS Hackathon, competing as Team Krsna, we wanted to tackle a very specific DevOps nightmare: how do you safely execute infrastructure commands on a local machine from a remote mobile device without opening your local firewall to the public internet?
Our answer was BotOps—a secure, local-first AI orchestration desktop application. To pull this off under a strict deadline, we divided and conquered. I took on the local backend and AI orchestration, while my teammate engineered the entire AWS cloud infrastructure.
The Vision and the Split The core philosophy of BotOps was zero-trust remote execution. We needed a system that could receive instructions from anywhere in the world but process them with the security and privacy of an air-gapped machine.
My teammate architected a brilliant serverless ingress pipeline in the cloud. They configured an Amazon API Gateway webhook to catch mobile commands and securely drop them into an Amazon SQS queue. This allowed our local machine to safely long-poll for instructions without ever exposing an inbound port. Later, they expanded this cloud footprint by integrating Amazon Bedrock to run efficiency evaluations on various foundation models, seamlessly extracting latency and token throughput metrics to log into a DynamoDB table for real-time telemetry.
While they were commanding the cloud, I was deep in the local backend building the brain of the operation. I developed the desktop UI using CustomTkinter and built the autonomous agent loop—which we named ChiefBot—using the Strands SDK. To ensure the bot could reason without sending sensitive data back to the cloud, I integrated Ollama to run Llama 3 directly on the local machine. Finally, I wrapped the entire execution environment in the Cedar policy language, creating a strict security guardrail that evaluated every proposed terminal command before it was allowed to run.
The Hackathon Hurdles No hackathon is complete without a chaotic technical roadblock. For me, it was an all-out war with Windows environment permissions. While trying to script the Ollama installation for the local LLM, I was repeatedly hit with  Exit Code 5  access denied errors, blocked  winget  commands, and terminal path nightmares that refused to recognize the binary. After hours of troubleshooting, I abandoned the automated installers entirely, manually downloaded the compiled  .exe  zip, and extracted it directly into our project directory just to force the local brain online.
The Breakthrough That struggle made our final integration test incredibly rewarding. I will never forget watching the logs during our first successful end-to-end run.
We fired a simple JSON payload from a remote terminal to the API Gateway. We watched the AWS infrastructure instantly queue the message in SQS. Seconds later, my local Python background thread pulled the message from the cloud, and the BotOps UI sprang to life. ChiefBot processed the instruction through the local Llama 3 model, passed the proposed action through my Cedar security guardrails, and flawlessly executed the command on the local terminal—while simultaneously pinging my teammate's DynamoDB table with the efficiency metrics.
Building BotOps was a masterclass in distributed architecture. My teammate's robust AWS serverless pipeline perfectly complemented the secure, local-first Python backend I engineered. It was a sleep-deprived, high-stress sprint, but watching our cloud-to-edge bridge execute autonomously made every single bug worth it.
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article