Empowering Nutritional Choices with Local Food Intelligence

Case Study: Empowering Nutritional Choices with Local Food Intelligence

Client Overview

Healthy Basket is a health-focused technology company committed to helping users make informed dietary decisions. Their vision is to make food awareness more accessible by enabling users to scan any meal and receive instant insights on its nutritional content, health rating, and cultural relevance, starting with the rich and diverse culinary traditions of Kenya.

The Challenge

While developing their intelligent food scanner, Healthy Basket faced a significant limitation: most existing food recognition solutions failed to accurately identify local Kenyan foods. This posed a serious roadblock, especially for users relying on accurate food analysis to manage their diets or health conditions.

The challenge was clear: how can we accurately recognize and provide detailed insights about local Kenyan meals?

The Solution: A Two-Stage Intelligence System

To solve this, Healthy Basket partnered with CloudPlexo, an AWS advanced partner, to develop a custom food recognition pipeline, combining computer vision with large language model (LLM) reasoning:

1. Custom Vision Model for Local Recognition

A computer vision model was trained using Amazon SageMaker Notebooks with a diverse dataset of local Kenyan foods. This model performs the initial recognition step:

• Outputs the food name

• Assigns a confidence score based on its certainty in the recognition

For example; If the confidence score is above 90%, the result is deemed reliable. If it’s below 90%, it indicates uncertainty—possibly a non-Kenyan food or an ambiguous image.

2. LLM-Powered Reasoning and Enrichment

Next, the food name, confidence score, and image are passed to Amazon Bedrock, where a Claude Sonnet 3.7 LLM:

• Analyzes the image again (especially when confidence is low)

• Cross-checks and potentially updates the food name

• Enriches the result with:

  • A brief summary of the food

  • A nutrient overview

  • A verdict on whether the food is healthy or not

Architecture Overview

Here’s how the full workflow unfolds:

1. User Action: A user takes or uploads a picture of food.

2. API Gateway & Lambda: The image is sent through Amazon API Gateway, and routed by AWS Lambda to the appropriate model.

3. Model Inference: A computer vision model trained on Kenyan foods processes the image and returns a food name and confidence score.

4. Image Analysis by LLM:

5. Data Enrichment: The LLM provides:

   • Final food name

   • Confidence score

   • Food summary

   • Nutritional overview

   • Health rating

6. Data Storage:

   • The input image is stored in Amazon S3

   • The final result is stored in Amazon DynamoDB

7. DNS Management: Amazon Route 53 is used to manage application routing and domain access.

AWS Services Used

• Amazon API Gateway – Receives user image requests

• AWS Lambda – Orchestrates the request-routing process

• Amazon SageMaker Notebooks – Used for training the custom food recognition model

• Amazon Bedrock (Claude Sonnet 3.7) – Performs LLM-based enrichment and fallback analysis

• Amazon DynamoDB – Stores analyzed food data and results

• Amazon S3 – Stores user-uploaded food images

• Amazon Route 53 – Handles DNS and application routing

The Results

With this intelligent pipeline built by CloudPlexo, Healthy Basket now offers:

• Accurate food recognition for local Kenyan dishes

• Personalized insights for users focused on healthy eating

• A scalable backend powered by AWS AI/ML and serverless infrastructure

This innovative solution empowers users to understand their meals better, embrace cultural foods with confidence, and take control of their health.

Conclusion

By combining computer vision and LLM capabilities within a serverless AWS architecture, Healthy Basket turned a critical recognition gap into an opportunity for innovation.

CloudPlexo helps businesses design, build, and scale solutions on AWS with confidence, whether it’s food tech, fintech, or beyond. Let’s talk about what you’re building next.