refundly:AI live verification for food delivery refund claim

The description states that Refundly is an AI-powered live verification platform designed for developers building food delivery and fast-food applications.

• Developers can generate API keys to connect their apps securely.
• They configure webhooks so backend systems receive results automatically.
• To initiate a session, the developer sends:
  • An order number
  • An optional note describing the complaint
  • A reference image of the food (used only for AI understanding, not as evidence)
• The system generates a unique verification link sent to the customer.
• During the live session, an AI assistant asks unpredictable actions like:
  • Making an OK hand gesture
  • Rotating or moving the plate
  • Showing food from another angle
  • Opening/cutting the food
• These actions help confirm real-time capture and validate food condition.
• Results are sent via webhook to the delivery platform indicating approval, rejection, or manual review.

Inference The product is described as a developer tool, not a consumer-facing service. It integrates into existing platforms through APIs and webhooks.

The description states that Refundly was inspired by the need for more reliable evidence in refund claims due to AI-generated manipulation of photos and videos.

• Traditional methods (photos/videos) are unreliable because they can be edited or manipulated.
• The solution is an interactive, real-time verification process that makes fraud harder.
• It positions itself as a developer platform, not a direct customer-facing tool.
• The author claims it's more than a visual AI experiment—it’s a complete workflow for integrating into delivery apps.

Inference The positioning evolved from solving a specific problem (AI fraud in food delivery) to offering a scalable developer infrastructure that could be adopted by multiple platforms.

The description states that Refundly is built for developers building food delivery and fast-food applications.

• The target is not end-users but platform developers or internal engineering teams at delivery companies.
• It supports integration via API keys, session creation, webhook configuration, and dashboard access.
• No mention of direct customers (e.g., restaurants, drivers, or consumers).

Inference The ICP appears to be B2B SaaS developers working on logistics or food tech platforms, with a focus on fraud prevention.

Not evidenced.

The description does not state anything about pricing models, monetization strategies, or commercial terms. It only describes the technical functionality and developer integration flow.

The description states that Refundly was built using:

• Codex for planning and code generation
• GPT-5.6 for workflow reasoning and AI assistant behavior
• FastAPI, Firebase, React, Python

It also mentions:

• API key generation
• Webhook configuration
• Live video session handling
• Session lifecycle management
• Reference-image processing (not used as evidence)
• Developer dashboard and documentation

Inference The technical stack suggests a modern, scalable architecture suitable for developer tools. The use of Codex implies rapid prototyping and iterative development.

Not evidenced.

There is no mention of:

• Revenue
• Customers
• Usage metrics
• Product adoption
• Market traction
• Any form of user base or pilot program

The project was submitted as a hackathon entry, indicating it's in early-stage development.

Not evidenced.

The description does not reference any competitors, market size, or competitive landscape. It only describes the problem and solution without placing itself within a broader industry context.

• No commercial traction: The project is a hackathon prototype with no evidence of real-world usage.
• Unproven developer adoption: No indication that developers are interested in integrating this platform.
• Technical complexity assumptions: Requires integration into existing delivery systems via API and webhooks, which may be non-trivial.
• AI reliability concerns: While the system uses AI for verification, it does not claim absolute certainty—this is a limitation rather than a strength.
• Scalability risk: The platform must handle live sessions, webhook retries, and security controls—all of which are mentioned as future improvements.

1. What specific food delivery platforms or developers have expressed interest in integrating this tool?
2. How does the system differentiate between legitimate and fraudulent users during live verification?
3. Are there any known edge cases where the AI might fail to detect manipulation?
4. Has the team considered how to scale the live session infrastructure for high-volume use?
5. What is the current status of webhook delivery reliability, especially under failure conditions?
6. How will Refundly prevent abuse from malicious actors who attempt to circumvent verification?
7. Are there any planned partnerships or integrations with existing delivery platforms or support systems?