Productionize Our RAG API with CI/CD, Monitoring and Tested Rollback

Where we are today

We have a working internal RAG application: a FastAPI service, a background ingestion worker, PostgreSQL, and a managed vector database. The application runs on a manually configured virtual machine.

Developers deploy over SSH. Application secrets are managed inconsistently, logs are difficult to correlate, and rollback has not been tested.

We need an engineer to make this service operable by our small development team.

The handover we want

By the end of the project, a team member should be able to release a version through GitHub Actions, see whether the release is healthy, investigate a failed ingestion, and restore the previous application version using the runbook.

Our target platform is AWS ECS Fargate. Use Terraform for the agreed infrastructure and keep the design proportionate to an internal application with approximately 60 users.

Required delivery

- Container and configuration review for the API and worker.

- Separate staging and production configurations.

- Infrastructure as code for the agreed AWS resources.

- CI/CD with automated checks, image versioning, and production approval.

- Managed secret injection and least-privilege service access.

- Structured logs and request correlation.

- Dashboards for latency, errors, ingestion failures, and model usage where available.

- Actionable alerts delivered to our existing team channel.

- A demonstrated application rollback and a database restore exercise.

Acceptance is operational

We will verify the implementation through a release demonstration, an injected worker failure, an alert test, and a rollback performed by our team.

Load testing should measure the agreed representative workload, starting with 10 concurrent users. Document bottlenecks and provider limits rather than assuming all latency is under infrastructure control.

Our team retains ownership of application logic, retrieval quality, and data-source permissions. This engagement excludes a new RAG architecture, a database migration project, and ongoing on-call coverage.

Include an infrastructure cost estimate, architecture notes, source-controlled runbooks, one handover workshop, and 14 days of bug support. Cloud charges are separate from the project budget.

Please propose milestone payments tied to staging readiness, production readiness, and the completed recovery drills.