
Weekend Showcase Challenge: OpsArena
OpsArena is an AI-powered AWS incident simulator that lets developers practice investigating metrics, reading logs, choosing remediation, and learning from an engineering-style postmortem.
Vision & What It Does
Production incidents are difficult to learn from because the safest place to make a mistake is usually not production. OpsArena is an interactive cloud incident simulation arena where developers and cloud learners practice responding to realistic AWS failure modes without touching a real workload.
The experience starts with a simple question: production is down, what do you do next? A responder chooses a difficulty and environment, then receives an incident brief. The console shows a severity level, error rate, latency, database connections, compute pressure, concurrency, logs, a service topology, and a timeline of what has happened so far. The responder investigates the evidence, opens the panels that matter, asks the Incident Commander for a clue, and chooses a remediation from the playbook or writes a custom action in plain language.
OpsArena is more than a chatbot. The simulation engine maintains state for the incident. An action can lower error rates and restore health, improve one symptom while leaving the root cause in place, or make the outage worse. A timeout change, for example, can keep requests alive longer and increase pressure on a constrained dependency. A source-level fix can stabilize the system and end the incident. Every action changes the health bar, score, metrics, and incident timeline.

The application includes twelve scenarios across serverless, compute, data, database, network, and application incidents. Amazon Bedrock generates a fresh evidence-based incident brief from a safe mechanical playbook, while the deterministic engine keeps the possible consequences consistent. Bedrock also powers contextual commander answers, action evaluations, and a final post-incident review. The app has a fallback engine so the training loop remains usable when model access is temporarily unavailable.
How I Built It
I started with the smallest useful loop: create an incident, expose evidence, accept a decision, update simulated infrastructure, and produce a postmortem. That led to a single Lambda-backed application instead of a large frontend/backend monorepo. The Lambda serves the operations console as static assets and exposes the JSON API through a public Function URL. This kept deployment quick and left more time for the incident mechanics and interaction design.
The main design challenge was making AI feel like part of a system rather than a decorative chat panel. I solved that by splitting responsibility. Bedrock handles natural-language generation and explanation. A scenario playbook defines the legal action set and numerical consequences. The engine applies those consequences first, then Bedrock explains the tradeoff in the language of an incident commander. This means a generated answer can be helpful without being allowed to invent a state transition.
I also added validation around generated scenario JSON. If a model response is incomplete, the app falls back to the curated playbook. This is important for a training tool: a malformed response should not break the simulation or reveal an inconsistent result. The same approach is used for postmortems. The result is stored in DynamoDB, so a responder can refresh or share a completed review through a persistent URL.
AWS Services Used / Architecture Overview

The deployed application uses:
- AWS Lambda to serve the web console, API routes, simulation engine, profile calculations, and scheduled daily challenge handler.
- AWS Lambda Function URL as the public HTTPS entry point for the browser and JSON API.
- Amazon DynamoDB with on-demand billing to persist active simulations, generated scenario briefs, daily challenge state, scores, and anonymous responder progress.
- Amazon Bedrock with Amazon Nova Lite for scenario generation, incident commander guidance, action evaluation, and postmortem generation.
- Amazon EventBridge to invoke the daily challenge refresh without a user needing to click a button.
- Amazon CloudWatch Logs through the normal Lambda execution environment for operational visibility and debugging.
- AWS SAM to define and deploy the infrastructure as code, including the IAM permissions for DynamoDB and Bedrock.
The request path is Browser → Lambda Function URL → OpsArena Lambda. The Lambda reads and writes DynamoDB state, calls Bedrock when generative context is useful, and returns the updated simulation to the browser. The daily path is EventBridge → Lambda → DynamoDB, which ensures that the daily challenge exists before responders begin competing on it.
The app is intentionally simulated. It does not create EC2 instances, RDS databases, or broken production infrastructure. That keeps the project safe and inexpensive while still teaching real operational reasoning.
What I Learned
This build taught me that an AI feature becomes much stronger when it is given a bounded state machine. Prompting alone can produce an interesting explanation, but the combination of a deterministic engine and a generative narrator produces an experience that is both teachable and believable. I also learned how useful Bedrock Converse is for keeping one model interface behind several product features.
On the AWS side, I practiced using SAM to package a Node.js Lambda with dependencies, attach least-purpose permissions, expose a public Function URL, persist application state in DynamoDB, and trigger unattended work with EventBridge. I learned to treat model access and IAM as separate deployment concerns: the function needs the correct invocation permissions, and the account still needs model access enabled.
The most valuable product lesson was to make the consequences visible. A score alone is abstract. Showing health, latency, connections, logs, and the timeline change after a decision makes the lesson immediate. OpsArena turns a cloud failure into a short feedback loop: investigate, act, observe, explain, and try again.
Try OpsArena
The deployed application is available at:
The source and deployment configuration are available in:
https://github.com/gokulupadhyayguragain/OpsArena.git
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article