Input
A file is uploaded into a designated prefix in an Amazon S3 bucket.
AWS internship capstone · 2025
AWS Serverless Metadata Workflow
When a file is uploaded to Amazon S3, AWS Lambda reads its object metadata, converts it into a consistent record, and stores that record in Amazon DynamoDB.
- Core services
- S3, Lambda, DynamoDB
- Architecture
- Event-driven and serverless
- Original environment
- Isolated AWS internship project environment
- Public implementation
- AWS SAM, Python, tests, CI, alarms, runbooks
Public implementation tested
11 unit tests
100% statement and branch coverage
Overview
The workflow removes the manual step of inspecting every uploaded file before its metadata can be recorded or used elsewhere.
Processing
An S3 event invokes Lambda, which validates the event and reads object headers with HeadObject.
Output
A normalized metadata item is written to DynamoDB with a deterministic record identifier.
Architecture
Each service has one responsibility, which keeps the workflow small enough to troubleshoot and inexpensive to operate.
Storage and event source
Amazon S3
Stores uploaded files and emits object-created notifications only for the configured intake prefix.
Compute
AWS Lambda
Decodes the key, reads metadata, validates required values, and prepares a normalized item.
Persistence
Amazon DynamoDB
Stores one record per object identity and rejects duplicate conditional writes safely.
What I built during the internship
The original capstone covered the complete path from upload through storage and presentation.
- Connected S3 object-created events to a Lambda metadata extraction function.
- Defined and wrote structured metadata records to DynamoDB.
- Worked through IAM permissions, service configuration, logs, and failure investigation.
- Built a static frontend that explained the workflow and presented its results.
- Created a usage-based cost model using storage, requests, compute duration, reads, writes, and transfer.
- Documented the architecture and the operational assumptions behind it.
Public implementation
The public repository recreates the architecture without exposing internal Amazon files, accounts, or customer information.
Deployable infrastructure
template.yaml defines the S3 bucket, Lambda function, DynamoDB table, SQS failure destination, IAM permissions, log retention, and CloudWatch alarms.
Tested Lambda code
The Python function handles URL-encoded keys, versioned objects, deterministic record IDs, duplicate deliveries, and clear failure propagation.
Open the Lambda functionReliability and operations
S3 event notifications can be delivered more than once, so the workflow treats duplicate delivery as normal system behavior rather than an exceptional case.
1
Deterministic identity
The bucket, object key, and object version are combined into a stable record ID.
2
Conditional persistence
DynamoDB uses attribute_not_exists(RecordId) so a duplicate event cannot overwrite an existing record.
3
Retry and recovery
Lambda retries asynchronous failures and sends exhausted events to an encrypted SQS failure queue.
4
Monitoring
CloudWatch alarms cover function errors, throttling, and messages waiting in the failure queue.
Evidence
The public implementation is checked as code, infrastructure, and documentation rather than presented only as an architecture graphic.
11unit tests
100%statement coverage
100%branch coverage
1 commandlocal verification with
make checkRelated AWS writing
These articles provide the longer context behind the internship, cloud cost decisions, and the way I learned the platform.
AWS Cloud Support Internship: What I Actually Practiced
The training environment, troubleshooting process, capstone, and exact scope of the work.
Read article
AWS Free Tier: What Actually Costs Money
A practical checklist for finding active resources, setting alerts, and avoiding surprise charges.
Read article
Certifications and Continuous Learning
How certification study connected cloud theory to consoles, permissions, logs, and troubleshooting.
Read articleProject scope
Original work and public code are intentionally separated
The original project was completed in isolated AWS internship training and project environments. This repository is a public reconstruction of the architecture, not an export of internal Amazon material. It contains no production customer data and does not imply ownership of live customer systems or an enterprise on-call rotation.
Explore the rest of my work
My portfolio includes cloud support, full-stack development, AI systems, technical documentation, and open-source contributions.