SupportTrace

The company appears to be a solo project (1 person) submitted to the OpenAI 2026 hackathon, named SupportTrace. The author describes it as an incident resolution system that uses GPT-5.6 for investigation, requires human approval of one action, and enforces deterministic verification before closing tickets. It is built with Next.js, TypeScript, GPT-5.6, Firestore, Google Cloud Run, and related tools.

What changed: The project was submitted to a hackathon, indicating an early-stage prototype or proof-of-concept. No evidence of revenue, customers, or product-market fit beyond the author's own description.

Single most important open question: Is there any evidence that this system has been tested in real-world support environments, or that it can scale beyond a single developer’s prototype?

Analysis basis: Self-reported and unverified. All claims are from the project description supplied by the caller. No third-party verification, traction data, or financials are available.

The description states that SupportTrace is a system for managing support incidents. It processes tickets through four steps:

1. Gather evidence.
2. Use GPT-5.6 to investigate and draft one proposed action.
3. Require human approval of that one action.
4. Verify the outcome deterministically before closing the ticket.

Every decision is recorded and replayable, allowing reconstruction from raw evidence to final state.

Inference: The system is designed as a deterministic audit trail for incident resolution, not an autonomous agent. It enforces a strict human-in-the-loop model with machine-assisted investigation.

The author states that the product was built to address the problem of support teams drowning in tickets and the risks of "let the AI just handle it" — which leads to silent, unaccountable mistakes. The positioning is clear: an AI-powered investigation tool with a hard boundary between AI and human action.

Claim: The system is not meant to automate actions but to improve accountability and traceability in incident resolution.

The author also claims that the value lies in “an agent that can do less” — i.e., it’s more trustworthy when it cannot act autonomously.

Inference: This is a positioning shift from AI automation toward AI augmentation with strict human oversight, emphasizing safety and auditability over speed or autonomy.

The description does not name specific customers or personas. However, the author implies that the system targets support and operations teams who are overwhelmed by tickets and need accountability in their workflows.

Claim: The product is aimed at organizations with support ticketing systems where traceability and human oversight are critical.

Inference: The ICP likely includes mid-to-large-sized tech companies or SaaS providers managing high volumes of incident tickets, though no evidence supports this.

No business model or pricing information is provided in the description. The project is presented as a hackathon submission with no indication of monetization strategy, customer acquisition plans, or revenue streams.

Not evidenced: No evidence of any business model, pricing, or commercialization plans.

The system is built using:

• Framework: Next.js
• Language: TypeScript
• AI: GPT-5.6 (via OpenAI Responses API)
• Backend: Google Cloud Run, Firestore, Secret Manager
• Testing: Vitest, Playwright
• Validation: Zod schemas
• Collaboration tooling: Codex

The system enforces a strict separation between investigation and action via type-level and permission-level constraints.

Claim: The system is deterministic and replayable because all actions are pure functions with testable outputs.

Inference: The architecture suggests a strong emphasis on safety, traceability, and auditability, not scalability or performance optimization.

The project is described as a hackathon submission. There is no evidence of:

• Revenue
• Customers
• Product-market fit
• Usage metrics
• Deployment in production environments

Not evidenced: No traction or maturity indicators beyond the author’s own account.

No mention of competitors or market context is provided in the description. The author does not reference existing tools for incident management, AI-assisted support, or ticketing systems.

Not evidenced: No competitive landscape or positioning against existing solutions.

• Single-person team: The entire project was built by one person (Efthimios Fousekis), suggesting limited scalability or long-term maintenance.
• No real-world testing: The system is described as a hackathon prototype with no evidence of deployment or use in production.
• Unproven AI integration: GPT-5.6 is used, but there’s no data on accuracy, hallucination rates, or performance in real scenarios.
• Limited scope: The project focuses only on one workflow (incident resolution) and does not indicate plans for broader functionality.

Inference: The system may be too narrowly scoped to be viable beyond a prototype, and the lack of testing raises questions about its robustness.

1. Has this system been tested in any real-world support environment?
2. What is the expected human effort required for approval and verification steps?
3. How does the system handle edge cases or ambiguous evidence inputs?
4. Are there plans to integrate with existing ticketing systems (e.g., Jira, Zendesk)?
5. What are the performance and latency characteristics of the AI investigation step?
6. How is the deterministic verification layer validated in practice?