System Requirements Document (SRD)

Project Name: meta-assessment

1. Introduction

The meta-assessment project is an AI-powered backend platform designed to automate the condition assessment of rental machines returned by customers. XYZ, based in India, operates a subscription model for domestic appliances, and this project aims to streamline the evaluation process for returned machines. By leveraging computer vision (CV) and artificial intelligence (AI), the system will replace the current manual assessment process, ensuring consistency, scalability, and efficiency.

This document outlines the system requirements for the meta-assessment project, including functional and non-functional specifications, user personas, design concepts, and technical considerations.

2. System Overview

The meta-assessment system will enable end-users or field staff to capture 5–6 photographs of returned machines. Using AI and CV technologies, the system will automatically evaluate the condition of the machines and provide a composite condition grade based on a universal grading rubric: Excellent, Good, or Fair. The system will also generate annotated images, component-level breakdowns, and refurbishment recommendations.

Key features include:

• Assessment of external body condition (scratches, dents, cracks, stains, discoloration).
• Evaluation of accessories (presence/absence and condition of detachable components like taps, brackets, trays).
• Inspection of power cords (fraying, cuts, heat damage, safety assessment).
• Analysis of internal components and filters (discoloration, sediment buildup, deformation, leakage marks).
• Examination of sensors and electronics (corrosion, water ingress, loose connections).
• Real-time or near-real-time processing of 50,000–60,000 images per month for 8,000–10,000 machines across 4 SKU types.

3. Functional Requirements

• As a User, I should be able to upload 5–6 images of a returned machine for condition assessment.
• As a User, I should receive a composite condition grade (Excellent, Good, Fair) based on the universal grading rubric.
• As a User, I should be able to view annotated images highlighting detected issues (e.g., scratches, dents, corrosion).
• As a User, I should receive a detailed component-level breakdown of the machine's condition.
• As a User, I should receive refurbishment recommendations based on the machine's condition grade.
• As an Admin, I should be able to configure the grading rubric and thresholds for condition categories.
• As an Admin, I should be able to monitor system performance and processing metrics.
• As an Admin, I should be able to generate reports on machine condition trends and refurbishment statistics.
• As an Admin, I should be able to manually trigger selective re-run capability for SRD regeneration.

4. User Personas

1. End-User

• Role: Field staff or warehouse personnel responsible for capturing machine images and uploading them to the system.
• Goals: Quickly assess machine condition and receive actionable insights for refurbishment or accountability.
• Pain Points: Manual assessments are slow, inconsistent, and prone to human error.

2. Admin

• Role: Supervisors or managers overseeing the assessment process and system performance.
• Goals: Ensure the system operates efficiently, monitor trends, and generate reports for business insights.
• Pain Points: Difficulty in scaling manual processes and ensuring consistent grading across SKU types.

5. Visuals Colors and Theme

Color Palette:

The color palette for meta-assessment reflects professionalism, precision, and clarity, aligning with the project's focus on automated assessments.

• Background: #EAEFF2 (Soft Sky Gray)
• Surface: #FFFFFF (Pure White)
• Text: #2C3E50 (Deep Navy)
• Accent: #3498DB (Vivid Cerulean Blue)
• Muted Tones: #95A5A6 (Neutral Slate Gray)

This palette ensures a clean, modern interface that emphasizes clarity and ease of use.

6. Signature Design Concept

Interactive Machine Anatomy Dashboard

The homepage of the meta-assessment system will feature an Interactive Machine Anatomy Dashboard. Users will see a 3D model of a generic rental machine that dynamically highlights different components (e.g., external body, accessories, power cord, internal filters, sensors) as they hover over or click on sections.

Key features:

• Dynamic Highlighting: When users hover over a section (e.g., power cord), the corresponding area on the 3D model lights up and displays a brief description of what the system assesses.
• Annotated Image Carousel: Below the 3D model, users can scroll through sample annotated images showcasing detected issues like scratches, dents, or corrosion.
• Color Transitions: The background subtly shifts between shades of Cerulean Blue (#3498DB) and Neutral Slate Gray (#95A5A6) to create a calming yet professional atmosphere.
• Micro-Interactions: Clicking on a component triggers a smooth zoom-in animation, revealing detailed information about the assessment criteria for that part.
• Real-Time Grading Simulation: A demo feature allows users to upload a sample image and see how the system grades it in real-time.

This bold design concept ensures the homepage is both visually striking and functionally informative, leaving a lasting impression on users.

7. Non-Functional Requirements

• Performance: The system must process 50,000–60,000 images per month with real-time or near-real-time results.
• Scalability: The system should handle increasing volumes of returned machines without degradation in performance.
• Accuracy: The AI models must achieve at least 95% accuracy in detecting and grading machine conditions.
• Security: All uploaded images and assessment data must be securely stored and transmitted using encryption protocols.
• Availability: The system must maintain 99.9% uptime to ensure continuous operation.

8. Tech Stack

Backend:

• Python
• FastAPI

Database:

• MySQL or MariaDB (preferred for relational data)
• MongoDB (for unstructured image metadata)
• Alembic for database migrations

AI Models:

• GPT 5.2 for user-friendly responses
• Claude 4.5 Opas for academic or coding work
• Google Nano Banana for image generation

AI Tools:

• Litellm for LLM Routing
• Langchain

Local Orchestration:

• Docker
• docker-compose

Server-Side Orchestration:

• Kubernetes

Relevant AWS Services:

• Amazon Lookout for Vision
• Amazon Rekognition Custom Labels
• Amazon SageMaker
• Amazon A2I
• AWS Lambda
• Amazon S3
• Amazon QuickSight

9. Assumptions and Constraints

Assumptions:

• Users will provide clear, high-quality images of returned machines.
• The universal grading rubric will remain consistent across all SKU types.
• AWS services will be used for AI model deployment and data storage.

Constraints:

• Processing time for each machine must not exceed 5 minutes.
• The system must operate within the budget allocated for AWS services and infrastructure.

10. Glossary

• SKU: Stock Keeping Unit, a unique identifier for a product type.
• AI: Artificial Intelligence, used for automating machine condition assessment.
• CV: Computer Vision, a field of AI focused on image analysis.
• Annotated Images: Images with visual markers highlighting detected issues.
• Refurbishment Recommendation: Suggestions for repairing or reconditioning returned machines.

This document provides a comprehensive overview of the meta-assessment project requirements. Let me know if there are additional details you'd like to refine, XYZ!