SupportTrace Evidence Gate

The description states that SupportTrace Evidence Gate is a keyless RAG support console. It retrieves approved knowledge, composes deterministic drafts, and attaches exact source excerpts. It calculates a transparent citation-relevance heuristic, then emits one of three explicit decisions: ready_to_send, human_review_required, or blocked. It also blocks prompt injection before release.

It is described as a TypeScript application with:

• Fastify API
• React/Vite console
• Deterministic local embeddings
• In-process vector store
• Vitest coverage
• Docker files and CI

The system is exposed through a POST endpoint /api/support/draft and includes a judge-visible dashboard.

Inference The product appears to be a proof-of-concept or demo-level tool, not a production-ready SaaS offering. It was built for a hackathon and is described as deterministic, keyless, and local.

The author states that the system addresses a problem: “Support drafts can sound confident even when retrieval is weak.” The product aims to give operators more than fluent prose — it provides exact evidence, confidence scoring, and explicit routing logic.

It positions itself as a tool for evidence-based support workflows, where human review is not a fallback but an explicit product state. It also claims to block prompt injection and separate evidence quality from policy risk.

Inference The positioning implies a shift from generic AI-generated support to a controlled, traceable, and accountable workflow — likely targeting enterprises or teams managing sensitive customer interactions.

The description does not explicitly name target customers. However, it is implied that the system is designed for:

• Teams managing customer support workflows
• Organizations needing evidence-based decision-making in support
• Users who want to retain control over sensitive or high-risk cases

It is described as a support console, suggesting a B2B SaaS or internal tooling use case.

Inference The ICP likely includes enterprise support teams, compliance-focused organizations, or customer service platforms that require auditability and risk control in AI-generated responses.

The description does not contain any information about pricing, monetization, or business model. It is a self-reported demo project built for a hackathon.

Inference There is no evidence of a commercial model, revenue, or pricing structure beyond the author’s own account.

• Built with TypeScript
• Uses Fastify API
• Includes React/Vite console
• Implements deterministic local embeddings
• Employs an in-process vector store
• Has Vitest coverage and Docker files
• CI/CD pipeline is included
• The system is described as keyless, local, and deterministic

The author states that Codex with GPT-5.6 was used for design, implementation, testing, documentation, and privacy verification.

Inference The technical stack suggests a developer-focused tool, likely built for internal use or demonstration. It is not described as cloud-hosted or scalable beyond the demo environment.

The project is described as a hackathon submission, built during Build Week, and includes:

• Eight passing tests
• A keyless local judge path
• Public-safe demo
• Reproducible documentation

There is no evidence of:

• Customers
• Revenue
• Product adoption
• Live usage
• Production deployment

Inference The system is at a demo or prototype stage, with no indication of traction or real-world maturity.

The description does not mention competitors. However, it implies a role in the RAG and AI support workflow space, where tools like:

• LangChain
• LlamaIndex
• Rasa
• Zendesk AI
• Salesforce Einstein

are present. The product’s focus on evidence attachment, confidence scoring, and explicit routing may differentiate it from general-purpose RAG systems.

Inference It is positioned to compete in a niche where traceability, auditability, and risk control are critical — but no specific competitors are named or compared.

• The system is described as deterministic, keyless, and local, which may limit scalability or integration.
• It was built for a hackathon, with no evidence of production use or real-world testing.
• There is no mention of data privacy, compliance, or enterprise readiness.
• The threshold score (0.40) is described as a demo policy, not calibrated.
• No evidence of customer feedback, error rates, or performance metrics beyond tests.

Inference The product may be too early-stage for commercial deployment and lacks any indication of real-world validation or enterprise-readiness.

1. What is the actual use case or customer problem you are solving?
2. Have you tested this system on real support traffic or with real users?
3. How does the 0.40 confidence threshold translate into business impact or error costs?
4. Is there a plan to move beyond the demo environment (e.g., cloud deployment, integration)?
5. What is the long-term vision for this product — is it intended as an internal tool or a SaaS offering?
6. How do you plan to calibrate the confidence scoring with real-world data?