Most people on USAJOBS.gov only need jobs in their own profession, and the main site can be cumbersome for that. I built a static site generator, inspired by Hugo (which I used years ago), that publishes every open federal IT Specialist (2210) job as one fast table with keyword and vector similarity search.
A separate data pipeline cleans, formats and embeds the postings on a slow refresh, since federal jobs don’t need second-by-second updates. The generator pulls that data at build time and ships a plain static page: no server, no database, no AI model in the browser.
- JAMstack
- TypeScript
- React
- Vite
- Tailwind CSS
- Web Workers
- BM25 + vector similarity
- Python
- GitHub Pages
Open the job table →
GitOps agents that run on AWS Bedrock and answer slash commands in GitHub. A comment like /hirobot on an issue or pull request sends a webhook through API Gateway to a Python Lambda. The Lambda asks a Bedrock model and posts the answer back as a comment.
The Lambda’s IAM role is scoped to a single Bedrock model and to its own secrets and logs. A GitHub Actions workflow can deploy it through OIDC, so no AWS keys are stored in GitHub. The first agent, hirobot, proved the full loop and has since been retired.
- Python
- AWS Lambda
- API Gateway
- AWS Bedrock (Nova Micro)
- Secrets Manager
- IAM
- GitHub webhooks
- GitHub Actions (OIDC)
View the source on GitHub →
The repo the agents work in. Slash commands go in its issues and the agent replies there, so issue #1 shows the round trip. It will also hold the schemas, policies and code for the agents to analyze.
View the repo on GitHub →