ClaimSight

The description states that ClaimSight turns home evidence into policy-aware insurance claims. It supports:

• Upload of walkthrough videos or photos.
• Vision AI identifying possessions room-by-room.
• Conversational intake for users with no evidence.
• Deterministic pricing logic (ACV calculation).
• Policy PDF parsing to extract coverage details and show gaps.
• Export of claims in structured formats (CSV/PDF).
• Bilingual correspondence assistant.
• Human review console for complaint drafts.

It is described as a vertical slice that runs end-to-end, including no-evidence paths.

Inference The product is built for disaster victims who need to document lost items and submit claims efficiently.

ClaimSight positions itself as an AI assistant that makes insurance claims faster and more accurate by leveraging existing home evidence. It emphasizes:

• Speed: “Minutes, not weeks.”
• Cost-efficiency: “No 10–15% adjuster fees.”
• Defensibility: “Uncertainty stays visible”.
• Accessibility: Works even with no evidence.

The author describes the tool as a solution to a common problem in insurance claims — lack of documentation — and frames it as an alternative to public adjusters, who charge high fees.

Inference The positioning is focused on reducing friction and fraud risk in insurance claims, especially for victims with limited documentation.

The target customer is homeowners who have experienced a disaster (e.g., fire, flood) and need to file an insurance claim. The tool is designed to help those who:

• Have no receipts or documentation.
• Have some home evidence (video, photos).
• Are trauma-affected and need support navigating the claims process.

The ICP appears to be individuals in high-risk regions or those with limited access to public adjusters.

Inference The ICP is trauma-affected individuals in need of documentation support, not necessarily a business-to-business model yet.

There is no evidence of pricing or monetization strategy beyond the author’s own description. The project is described as a hackathon prototype with no revenue data or customer base.

Inference The pricing logic is deterministic and not reliant on generative AI, but no commercial model or pricing tiers are described.

ClaimSight was built using:

• Next.js (App Router, TypeScript)
• Supabase for anonymous job persistence
• Gemini 2.5 Flash for vision extraction, policy parsing, and conversational intake
• Codex as the primary engineering agent
• AI used for extraction and language; math stays auditable

The system is described as:

• Deterministic core with AI at edges.
• Resilient to offline use (demo runs without API keys).
• Built with a single schema that supports multiple input methods.
• Includes human review console for complaint drafts.

Inference The engineering approach is agentic and focused on minimizing risk through deterministic logic.

There is no evidence of traction, revenue, or customer adoption. The project was built in one week as part of a hackathon and is not described as having launched or scaled beyond that.

Inference The tool is at a prototype stage and has not been validated in the market beyond its own author’s account.

There are no references to existing competitors. The author states that “nobody had built the tool that turns that evidence into a claim,” suggesting a lack of direct competition, but this is self-reported.

Inference The competitive landscape is unclear, and there is no indication of existing solutions in the insurance claims space for this use case.

Key risks include:

• Legal risk: AI hallucination could lead to fraudulent claims.
• Scalability risk: The tool is described as a prototype built in one week with a single developer.
• Trust risk: The product must be defensible and auditable, especially in high-stakes insurance contexts.
• Market risk: No evidence of customer validation or market traction.

Inference The tool has not been tested at scale or validated by users beyond its creator.

1. What is the actual legal and regulatory risk of AI-generated claims in insurance?
2. How does the system handle edge cases where evidence is ambiguous or incomplete?
3. Has there been any user testing or feedback from trauma-affected individuals?
4. What are the plans for scaling beyond a single developer and hackathon prototype?
5. Are there any partnerships or pilot programs with insurers or adjusters?
6. How is the human review console integrated into the broader workflow?