System Architecture

← Back to Journal

AI Trip Wildlife Project Architecture

A serverless, event-driven system that transforms field uploads into an automated travel journal using AWS services.

📋 System Overview

This is an automated travel journal built on AWS serverless infrastructure with OpenAI integration for intelligent audio transcription validation. The system processes photos, audio recordings, and text notes uploaded to S3, automatically generating timestamped blog posts served via CloudFront.

Core Workflow:
  • Audio Processing: Amazon Transcribe converts voice recordings to text, OpenAI validates and cleans transcriptions
  • Image Processing: Amazon Rekognition detects wildlife, objects, and scenes in photos
  • Content Assembly: Hourly batch job aggregates pending uploads into unified blog posts with metadata stored in DynamoDB

🏗️ Architecture Diagram

AI Trip Wildlife Project Architecture Diagram

Click diagram to enlarge

× AI Trip Wildlife Project Architecture Diagram

☁️ Services & APIs

🪣 Amazon S3

Two buckets: one for raw uploads, one for processed website content. S3 event notifications trigger Lambda functions.

⚡ AWS Lambda

Five serverless functions: start transcription, validate transcription, process photos, build blog posts, and hourly batch creator for aggregating pending content.

🤖 OpenAI API

GPT models for validating and cleaning Amazon Transcribe output to improve transcription accuracy.

🗣️ Amazon Transcribe

Converts voice recordings (.m4a files) into text transcriptions automatically.

🖼️ Amazon Rekognition

Detects objects, scenes, and wildlife in photos, generating automatic tags and labels.

🗄️ Amazon DynamoDB

NoSQL database storing metadata for each upload: timestamps, transcriptions, image labels, and processing status.

⏰ Amazon EventBridge

Scheduled events triggering hourly batch blog creation to aggregate pending content into unified posts.

🌐 Amazon CloudFront

Global CDN serving the static website with low latency from edge locations worldwide.

🔧 AWS SAM

Infrastructure as Code (IaC) for deploying Lambda functions, permissions, and event triggers.

🔐 AWS IAM

Fine-grained permissions for Lambda execution roles, S3 access, and service-to-service communication.

🔄 End-to-End Workflow

File Upload to S3

User uploads files to the source S3 bucket from mobile device or computer. Files are organized by timestamp and type (photos/, notes/, audio/).

S3 Event Triggers Lambda

S3 sends event notifications to Lambda functions based on file suffix:

  • .m4a files → trigger FnStartTranscribe
  • .jpg files → trigger FnProcessPhotos
  • .txt files → trigger FnBuildBlogPost

Audio Transcription

FnStartTranscribe starts an Amazon Transcribe job. When complete, FnValidateTranscription is triggered to download the transcript, store it in DynamoDB, and mark the audio as processed.

Photo Processing

FnProcessPhotos uses Amazon Rekognition to detect objects, scenes, and landmarks. It generates web-optimized thumbnails, stores labels in DynamoDB, and saves metadata for later blog assembly.

Blog Post Generation

FnBuildBlogPost queries DynamoDB for all entries matching the timestamp, assembles notes, transcriptions, and photos into an HTML post, and uploads it to the site bucket.

Index Update

The blog post is added to site/blog/index.json, which contains metadata for all posts (timestamps, S3 keys). This index is read by blog.html to display entries.

CloudFront Invalidation

Lambda creates a CloudFront invalidation for updated paths, ensuring users see fresh content immediately without cache delays.

User Views Journal

Users access blog.html via CloudFront. JavaScript fetches index.json, loads individual post HTML files, and renders them as magazine-style cards with photos, transcriptions, and notes.