Job Market Tracker
Ranks the technologies entry-level software postings actually require, and flags the ones I don't have evidence for yet.
- When
- Oct 2026 – Nov 2026
- Status
- Live
- Role
- Solo
- TypeScript
- React
- AWS Lambda
- DynamoDB
- SQS
- AWS CDK
Problem
I’m a new grad applying to software jobs every day, and I wanted to decide what to learn next from data instead of guesses. So I saved the postings I applied to and counted which technologies they ask for. The biggest gaps turned out to be AWS and TypeScript, so I built this with them.
What it does
- Dashboard ranks technologies by how many postings firmly require them. “Java, Python, or Go” doesn’t require Go, so mentions inside either-or lists are counted separately.
- Roles I can’t take are excluded. Postings that require citizenship or a security clearance are filtered out.
- Check a posting: paste any job description to see which technologies it asks for and which I have evidence for. It runs entirely in the browser.
Architecture
The dashboard is a React app on GitHub Pages that calls a read-only API (API Gateway → Lambda → DynamoDB). Saved postings are synced to a private S3 bucket; object events go to an SQS queue that batches uploads for 60 seconds and triggers an Ingest Lambda, which recomputes every count. Failed runs land in a dead-letter queue with a CloudWatch alarm that emails me. All AWS infrastructure is defined with CDK.
Design decisions
- Recompute everything on each upload. With 68 postings a full recompute takes under 4 seconds, and it makes every run idempotent, so retries are always safe.
- The posting check runs client-side. It reuses the same
corepackage as the backend, so the dashboard and the checker can’t disagree, and pasted text never leaves the page. - Costs near zero by design: on-demand DynamoDB, no VPC or NAT gateway, API throttled to 10 req/s, two-week log retention.
- Least privilege, enforced by tests. Infrastructure tests fail if the API Lambda gains write access or a bucket becomes public.
Results
| Metric | Result |
|---|---|
| Postings analyzed | 68 (5 more filtered out) |
| Full recompute | 1.0–3.9 s, at most 126 MB memory |
| API latency (warm) | 175 ms median, 276 ms p95 |
| Lighthouse performance | 99 mobile, 100 desktop |
| Tests | 45 Vitest tests across all four packages |