SLO Guardian

What the company appears to be: SLO Guardian is a self-reported system designed to monitor distributed systems for service level objective (SLO) violations, explain root causes using trace data, propose AI-generated fixes, and simulate interventions before human approval. It is built as an incident response tool that keeps AI suggestions within a controlled sandbox, with final decision-making remaining human.

What changed: The project was submitted to the OpenAI 2026 hackathon by one developer (Divya Kamila), indicating it is a prototype or proof-of-concept. No prior version or commercial product exists beyond this submission.

Single most important open question: Is there any evidence of real-world deployment, customer feedback, or traction beyond the author’s own test environment?

The description states that SLO Guardian monitors distributed systems for SLO violations and explains why a system is approaching failure. It uses trace data from tools like Jaeger and OpenTelemetry to identify issues.

It proposes interventions via GPT-5.6, but these are filtered through an internal safety layer that tests and rejects unsafe or unproven suggestions before they reach a human reviewer.

The AI never directly acts on the system — only suggests. Human approval is required for any fix to be applied, and all fixes are tested in a safe practice run first.

Inference: The product appears to be an incident response assistant that integrates AI with existing observability platforms (Jaeger, OpenTelemetry) and enforces a strict human-in-the-loop model.

The author claims SLO Guardian is built around the principle: “GPT proposes. The rules decide. The human approves.”

This positioning emphasizes trust in human judgment over autonomous AI action, particularly in high-stakes environments like system failures during peak traffic.

It positions itself as a tool that reduces stress and ambiguity during incidents by offering clear explanations and vetted options — not as an automated remediation engine.

Inference: The product is positioned as a safety net for SREs or DevOps teams who want AI assistance without relinquishing control. It does not claim to replace humans, but rather to augment their decision-making.

The description states that the system was tested on a small online store with five services (checkout, inventory, pricing, and recommendations). This suggests an early-stage use case involving small to medium-sized distributed systems.

It is implied that the primary users are SREs or DevOps engineers working in environments where service reliability is critical but not yet fully matured.

Inference: The target customer likely includes engineering teams managing complex, distributed applications who need better incident context and safer troubleshooting workflows.

There is no evidence of pricing, licensing, or monetization strategy in the description. The project is presented as a hackathon submission with no indication of commercial intent or business model.

Not evidenced: No mention of revenue streams, subscription models, or any commercial framework.

The system uses:

• Codex, Docker, FastAPI, GPT-5.6, Jaeger, OpenTelemetry, Python, React, TypeScript
• A dashboard for human review
• An internal safety layer that checks and tests AI suggestions
• Trace data integration from Jaeger and OpenTelemetry
• A practice store to simulate failures

The system is described as being built with a focus on safety — e.g., no AI touches the real system, all fixes are tested in isolation.

Inference: The technical stack suggests a lightweight, developer-oriented tool that integrates with existing observability infrastructure. It’s not a full SRE platform but rather a focused incident response assistant.

The only evidence of traction is from the author's own test environment:

• Checkout recovery time dropped from 1,163 ms to 382 ms
• Zero critical requests were dropped during testing
• Every fix was tested in a safe practice run

There is no mention of:

• Customers
• Deployments in production
• Real-world usage
• Feedback or adoption metrics

Not evidenced: No evidence of real-world deployment, customer base, or performance outside the test environment.

The description does not reference competitors. However, based on its functionality — monitoring SLOs, explaining failures, and suggesting fixes using AI — it aligns with a category of tools that help with incident response, root cause analysis, and observability in distributed systems.

It is distinct from fully autonomous remediation systems by design (i.e., it does not act without human approval).

Inference: SLO Guardian operates in the space of AI-assisted incident response or observability tools. It may compete with platforms that offer similar capabilities but lack its strict human-in-the-loop architecture.

• No real-world deployment: The system is only tested in a controlled environment.
• Unproven scalability: No evidence of how it would scale to larger or more complex systems.
• AI dependency risk: Reliance on GPT-5.6 may be fragile if access or performance changes.
• Limited scope: Only tested on a small, synthetic system; no indication of integration with real-world tools or workflows.
• No commercial viability: No evidence of monetization, market fit, or customer traction.

Inference: The product is currently a prototype. Its long-term viability depends on whether it can be extended beyond the test environment and gain traction in real-world SRE teams.

1. What are the specific use cases where this system has been tested outside of the hackathon?
2. How does the safety layer handle edge cases or unexpected inputs from GPT-5.6?
3. Has the team considered integrating with existing SRE toolchains (e.g., PagerDuty, Splunk)?
4. What is the plan for moving from a test environment to production use?
5. Are there any known limitations in terms of system complexity or scale that would prevent adoption?