meta-assessment
F(x) Data Labs Pvt LtdUse Case: Automated Rental Machine Condition Assessment The customer operates a subscription model where domestic appliances are deployed at customer homes. When machines are returned (cancellation, upgrade, relocation), they need to assess the condition to determine refurbishment eligibility, required repairs, and damage accountability. Current state: Manual assessment by warehouse staff — slow, inconsistent, and doesn't scale. What they need: An AI/CV system where end-users or field staff capture 5-6 photographs of the returned machine, and the system automatically evaluates: External body condition — scratches, dents, cracks, stains, discoloration Accessories — presence/absence and condition of detachable components (taps, brackets, trays) Power cord — fraying, cuts, heat damage, safety assessment Internal components & filters — discoloration, sediment buildup, deformation, leakage marks Sensors & electronics — corrosion, water ingress, loose connections The system should output a composite condition grade per machine with component-level breakdown, annotated images, and a refurbishment recommendation. Scale: 8,000 — 10,000 machines returned per month 5-6 images per machine (~50,000-60,000 images/month) 4 SKU types to assess Real-time or near-real-time processing required Relevant AWS services: Amazon Lookout for Vision, Amazon Rekognition Custom Labels, Amazon SageMaker, Amazon A2I, AWS Lambda, Amazon S3, Amazon QuickSight.
Comments (0)
Sign in to leave a comment
Project Tasks22
Implement S3 Image Storage Service
As a user I want to be able to use a backend service for reliable image storage at scale. Set up AWS S3 bucket configuration, IAM policies, pre-signed URL generation, and a reusable service layer for uploading, retrieving, and managing machine assessment images. Supports 50,000-60,000 images/month.
Implement MongoDB Metadata Service
As a user I want to be able to have assessment data reliably stored and retrievable. Design and implement MongoDB schema and service layer for storing image metadata, assessment records, grading results, annotated image URLs, and component scores per machine submission at scale.
Implement AI Grading Pipeline
As a user I want to be able to use an AI backend pipeline to automatically grade machine condition from uploaded images. Integrate Amazon Rekognition Custom Labels and Lookout for Vision to detect defects per component, map to Excellent/Good/Fair grades, output annotated images and composite score. Expose via FastAPI.
Implement Theme & Structure
As a user I want to be able to use a frontend that reflects the meta-assessment brand identity. Implement the global color palette (#EAEFF2 background, #FFFFFF surface, #2C3E50 text, #3498DB accent, #95A5A6 muted), typography (Inter font family), spacing tokens, and shared layout structure across all pages. Remove any scaffold pages not required by the user flow. All references to company names must use 'XYZ' only. This task is a prerequisite for all page-level UI tasks.
Implement Rubric Config API
As an admin I want to be able to use a backend API to configure grading rubric thresholds. FastAPI CRUD endpoints for rubric configuration per SKU type and component (body, accessories, power cord, filters, sensors), stored in MySQL.
Implement Lambda Async Orchestration
As a user I want to be able to trigger AI assessment processing automatically upon upload. Configure AWS Lambda functions to orchestrate asynchronous pipeline execution: fan-out to Rekognition/SageMaker upon S3 trigger, aggregate component results, and write final grades to MongoDB.
Implement A2I Human Review Integration
As an admin I want to be able to route low-confidence assessments to human reviewers. Integrate Amazon A2I to flag borderline machine grades, present review tasks to warehouse staff, collect annotations, and feed corrections back into the grading pipeline to improve model accuracy.
Implement Job Status Polling API
As a user I want to be able to use a backend API to poll for assessment job processing status. FastAPI endpoint returns job state (pending, processing, complete, failed) so the frontend can display real-time progress while the AI pipeline processes images asynchronously.
Implement Image Annotation Service
As a user I want to be able to view annotated images highlighting detected defects. Build a service that generates bounding-box overlays on machine images per component (body, accessories, power cord, filters, sensors), stores annotated images to S3, and saves URLs to MongoDB.
Implement Dashboard Metrics API
As an admin I want to be able to use a backend API to retrieve system performance metrics. FastAPI endpoints aggregate data from MySQL/MongoDB: machines processed per month, AI accuracy rates, grade distributions, and processing time stats for the Dashboard page.
Build Landing Page
As a user I want to be able to use a frontend Landing Page that serves as the entry point for end-users. Implement based on existing JSX design (v3). Includes Interactive Machine Anatomy Dashboard, annotated image carousel, real-time grading demo, feature highlights, and CTA to New Assessment. All references to company name must use 'XYZ' only. Links to Login and New Assessment pages.
Implement Results Retrieval API
As a user I want to be able to use a backend API to retrieve assessment results. FastAPI endpoint fetches composite grade, annotated image URLs from S3, component-level breakdown scores, and refurbishment recommendations from MongoDB by job ID.
Build Settings Page
As an admin I want to be able to use a frontend Settings Page to configure grading rubric thresholds. Implement based on existing JSX design (v3). Allows Admins to configure grading rubric thresholds for Excellent, Good, and Fair per SKU type and component. Changes persist to backend via Rubric Config API. Uses XYZ branding only.
Build Dashboard Page
As an admin I want to be able to use a frontend Dashboard Page to monitor system metrics. Implement based on existing JSX design (v4). Displays machine assessment metrics (volumes, accuracy, grade distribution), KPI cards, charts, and navigation to Reports and Settings. Entry point post-login for Admins. Uses XYZ branding only.
Implement Assessment Submission API
As a user I want to be able to use a backend API to submit machine images for AI-powered condition assessment. FastAPI endpoint accepts 5-6 images and machine metadata, stores images to S3, persists metadata to MongoDB, triggers AI grading pipeline asynchronously via Lambda, and returns a job ID for polling.
Build Login Page
As a user I want to be able to use a frontend Login Page to authenticate into the system. Implement based on existing JSX design (v4). Supports Admin authentication with form validation, error states styled to meta-assessment theme using XYZ branding only. Redirects to Dashboard on success.
Build New Assessment Page
As a user I want to be able to use a frontend New Assessment Page to initiate a machine condition assessment job. Implement based on existing JSX design (v4). Allows end-users to select SKU type and enter metadata. Navigated to from Landing CTA, links forward to Upload page. Uses XYZ branding only.
Implement QuickSight Reports Integration
As an admin I want to be able to view rich analytics dashboards and export reports. Connect assessment data from MySQL/MongoDB to AWS QuickSight: configure datasets, build analyses for grade distribution and refurbishment trends, and embed report URLs in the Reports page.
Build Results Page
As a user I want to be able to use a frontend Results Page to view the AI-generated condition assessment. Implement based on existing JSX design (v4). Displays composite grade (Excellent/Good/Fair), annotated images highlighting detected issues, component-level breakdown, and refurbishment recommendations. Navigated to after analysis completes. Uses XYZ branding only.
Implement Reports Generation API
As an admin I want to be able to use a backend API to generate and export reports. FastAPI endpoints aggregate assessment data from MySQL/MongoDB, support date range filtering, return exportable CSV/PDF, and integrate with Amazon QuickSight for advanced analytics.
Build Upload Page
As a user I want to be able to use a frontend Upload Page to capture and submit machine images for assessment. Implement based on existing JSX design (v4). Supports drag-and-drop upload of 5-6 machine images, component coverage checker, image preview grid, progress indicators, and submission to backend. On submission navigates to Results page. Uses XYZ branding only.
Build Reports Page
As an admin I want to be able to use a frontend Reports Page to view and export machine condition trends. Implement based on existing JSX design (v3). Displays machine condition trends, refurbishment statistics, aggregated data tables and charts, with CSV/PDF export functionality. Navigated to from Dashboard. Uses XYZ branding only.
No comments yet. Be the first!