AI-Powered Cloud Inference Pipeline Development

Job ID: 39239568

Budget: ₹12,500 – ₹37,500 INR

Project Title:
Cloud-Based AI Pipeline (YOLO + OCR) for Extracting Data from Dynamic Screens via Image Input

Description:
We are developing a system that captures images of dashboards/screens and extracts key labeled values (e.g., “Label AB,” “Label XY”) using a camera.
Our internal team handles the hardware/camera system and uploads images to the cloud.

We are looking for an experienced AI/ML + Azure cloud developer to build the cloud-side inference pipeline using computer vision + OCR techniques. The entire system must be Dockerized and deployed on Azure.

Scope of Work:
☁️ 1. AI/ML Inference Pipeline on Azure
Receive uploaded screen images from Azure Blob or an API

Preprocess images for:

Blur detection

Glare/reflection handling

Shadow correction

Screen cutoff validation

Use object detection (e.g., YOLO) to detect key screen regions ("Label AB", "Label CD", etc.)

Use OCR to extract numeric or alphanumeric values next to those labels

Match labels → values dynamically (layout agnostic)

Return structured JSON output including:

Detected values

Confidence scores

Any errors or quality issues flagged

? 2. Error & Validation Handling
Flag and tag issues such as:

Image is too blurry

Value not detected

Label missing

Low OCR confidence

Screen not fully visible

Provide this metadata in a clean JSON format

Example:

json
Copy
Edit
{
"label_ab": {
"value": "97",
"confidence": 0.91
},
"label_cd": {
"value": null,
"error": "Label not detected"
},
"image_quality": {
"sharpness_score": 28.5,
"glare_detected": true,
"issues": ["Blurry", "Cutoff at top"]
}
}
? 3. API Development (for mobile app integration)
REST API endpoints to:

Receive an image from the device

Return extracted values + validation flags

(Optional) retrieve previous results by ID or timestamp

API must include image error feedback that our mobile app team can consume and display to users during or after upload.

? 4. Dockerization & Deployment
Package the full pipeline into a Docker container

Deploy to Azure Container Instance or similar

Provide documentation for:

Building and running the container

Updating the model or logic

Environment variables and settings

? 5. Feedback Loop + Retraining Folder Setup
Save flagged/failed cases in a retraining-friendly folder structure

bash
Copy
Edit
/flagged_cases/
/retrain_data/
(Optional) provide a script or structure for retraining the model with new labeled data later

Support manual model upgrading (drop-in new model + rebuild Docker)

?️ Tech Stack Required:
Python (preferred)

OpenCV, PyTorch or TensorFlow

YOLOv5/YOLOv8 or similar detection model

OCR: EasyOCR, Tesseract, or PaddleOCR

Azure Blob Storage, Azure Functions or HTTP Trigger

Azure Container Instance (ACI) or App Service

REST API: FastAPI or Flask

Docker

✅ What We Provide:
Sample labeled screen images (2 dashboard layouts)

Base model checkpoint (if needed)

Azure environment + access

Input/output format specs

Our team handles image capture and uploads

? Deliverables:
Dockerized AI pipeline

Deployed on Azure (ACI preferred)

REST API with:

JSON outputs for detected labels + values

Error tagging + image quality flags

Folder structure + simple script for future retraining

Clear deployment + update documentation

✨ Nice to Have (Bonus):
Experience with model versioning

CI/CD familiarity (Docker push → deploy flow)

Ability to collaborate on feedback-based model improvement