Support Gap Radar

The description states that Support Gap Radar:

• Takes resolved support tickets (in CSV format) and current documentation (Markdown, text, PDF, Word)
• Uses OpenAI embeddings to group semantically similar questions
• Applies GPT-5.6 for structured coverage audit labeling (covered, partial, missing, contradiction)
• Visualizes knowledge gaps in a radar chart
• Drafts editable Markdown patches based on evidence
• Re-tests historical questions against updated documentation and patches
• Blocks drafting if documentation and support outcomes disagree, requiring human policy decision

The product is described as working with any support platform that exports tickets (e.g., Zendesk, Salesforce, Front), and it is built using React 19, Vite, Netlify Functions, and OpenAI APIs.

Inference This is a proof-of-concept or early-stage product built as part of a hackathon; no evidence of commercial deployment or customer adoption exists in the description.

The author states:

• The tool addresses a gap in support systems: documentation and tickets live separately, so teams can’t prove which help articles are incomplete or outdated.
• It is not a generic summarizer but a reusable loop: detect, patch, replay, review.
• It works as an intelligence layer above existing support platforms.

Inference The positioning implies a shift from reactive summarization to proactive, evidence-based documentation improvement — but this is not validated by real-world usage or feedback.

The description states:

• The tool targets support teams and knowledge teams within companies.
• It is designed for organizations that use support ticketing systems (Zendesk, Salesforce, Front) and maintain documentation in various formats.
• It supports a loop of detecting gaps, patching them, and replaying results.

Inference The ICP is likely large enterprises or SaaS companies with mature support operations and documentation systems — but no evidence of actual customers or use cases is provided.

Not evidenced.

The author states:

• Built with React 19 and Vite
• Uses Netlify Functions for server-side OpenAI calls
• Leverages text-embedding-3-small for semantic clustering
• Uses deterministic k-means and cosine similarity for repeatable grouping
• GPT-5.6 is used with structured outputs validated via Zod
• Supports browser parsing of CSV, Markdown, PDF, DOCX
• Includes editable Markdown export, contradiction blocking, and Knowledge Replay

Inference The technical architecture is described as robust enough to support a demo with reproducible AI behavior, but no evidence of production-grade scalability or reliability is provided.

Not evidenced.

Inference The product is at an early stage — a working prototype or proof of concept — but no evidence of traction or commercial maturity is present.

Not evidenced.

Inference The product appears to address a gap in existing tools but lacks any evidence of how it compares to or differentiates from other solutions in the marketplace.

• No commercial traction: The project is described as a hackathon submission with no revenue, customers, or adoption.
• Unverified AI claims: GPT-5.6 is used for structured outputs, but there’s no evidence of real-world performance or validation beyond a smoke test.
• Limited scope: The tool works on CSV and document formats, but lacks integration with actual support platforms in production.
• Self-reported maturity: The description implies the product is functional, but no independent verification or user feedback exists.

1. What is the actual process for integrating this with support platforms like Zendesk or Salesforce?
2. How does the tool handle conflicting documentation and support outcomes in practice?
3. Are there any real-world use cases or pilot programs beyond the hackathon demo?
4. What are the limitations of the current AI model outputs, especially around contradiction detection?
5. How is the tool validated for accuracy and consistency across different types of documentation?