MittiGuard

The description states that MittiGuard is a system that turns uncertain farm-input requests into evidence work. It records field details, crop stage, farmer’s story, previous input outcomes, soil reports, optional images, and weather context.

It uses Amazon Nova Pro through Amazon Bedrock to generate an editable evidence draft but does not recommend products or doses. If critical data is missing, the invoice is paused and tasks are created for human review.

The system includes:

• A server-side policy engine
• Field Memory
• Audit trail
• Mobile Field Capture links
• Evidence Debt detection (to block repeat sales)
• A Safety Bench for replaying controls

It is built with Node.js, JavaScript, HTML, CSS, and deployed on AWS Lightsail.

Not evidenced No actual product, revenue, or customer data. The system is described as a demo, not a live tool.

The author states that the inspiration was to avoid building a disease-diagnosis app that gives confident answers — which they felt would be wrong in cases of incomplete information.

Instead, MittiGuard is positioned around the idea of “not selling blind,” where evidence must be sufficient before any sale can proceed.

It claims to separate AI’s role in organizing information from its role in making decisions. The system is designed so that:

• AI helps draft evidence
• Policy engine controls sales
• No single model response leads to a sale

Inference The positioning implies a shift from reactive, data-driven decision-making to a more cautious, evidence-based workflow — but this is not validated by real-world adoption or feedback.

The description states that MittiGuard targets agri-input dealers who face pressure to give quick answers when farmers present unclear crop issues.

It also mentions that the next step would involve testing with “real dealers, field workers, and extension teams.”

Not evidenced No actual customer list, user roles, or market segmentation. The target is inferred from the use case described.

The description does not state any pricing model or business model beyond the idea of preventing premature sales.

It mentions that the system would need to be integrated into real dealer workflows and that it would require proper user roles, secure authentication, and durable databases for production deployment.

Not evidenced No revenue streams, pricing tiers, or monetization strategy are described. The project is presented as a demo with no commercial implementation.

The system is built using:

• Node.js
• JavaScript
• HTML/CSS
• AWS Lightsail
• Amazon Bedrock (Nova Pro)
• Codex and GPT-5.6 for development
• Docker, Caddy, JSON, HMAC-SHA256

It includes:

• Server-side policy engine
• Mobile Field Capture links with bounded observation storage
• Evidence Debt detection
• Audit trail
• Safety Bench for replaying controls

Inference The use of AWS and AI tools suggests a modern tech stack, but no production-grade infrastructure or scalability data is provided.

The project is described as a “synthetic jury demo,” not a production system. It includes:

• A public live demo
• Bypass-proof UI and server-side validation
• A 45-check Safety Bench

It has no real-world users, customers, or adoption data.

Not evidenced No traction, revenue, or user feedback is reported. The project is self-reported as a hackathon submission with no evidence of real-world usage.

The description does not mention any competitors or existing solutions in the agri-input or agricultural diagnostics space.

It implies that current tools may be too confident in their outputs, leading to ineffective or unsafe recommendations — but this is not substantiated by market data or competitive analysis.

Not evidenced No competitive landscape or differentiation strategy is provided.

• The project is described as a demo, not a production system.
• No evidence of real-world adoption or user feedback.
• The system relies on a single developer and lacks team structure or scalability.
• It uses AI tools like GPT-5.6 and Codex for development — which may not reflect long-term technical strategy.
• There is no mention of data privacy, consent, or regulatory compliance in the described workflow.

Inference The lack of real-world testing and user feedback raises questions about whether the workflow would be practical or accepted by dealers or extension teams.

1. What specific feedback have you received from agri-input dealers or extension teams during the demo phase?
2. How does the system handle edge cases, such as when a dealer intentionally bypasses a required task?
3. Are there any plans to integrate with existing dealer management systems or agricultural platforms?
4. What is the expected cost of deploying this system at scale?
5. How do you plan to onboard and train dealers on using the system?
6. Have you considered how to handle data retention, consent, and privacy for field observations?