hiring

DevOps Internship Assignment

Overview

In this assignment, you will deploy the quickstart project across multiple VMs running in a private subnet, wire the workers together over remote procedure calls (RPC), and expose model inference through a JSON HTTP API.

The quickstart project ships a tiny SLM along with a cross-language worker setup (Python + TypeScript). Because the model is small, the whole thing fits comfortably inside the free tier of either AWS or GCP — no GPU instance required.

Prerequisites

• Sign up on Google Cloud (you get $300 in free credits) or AWS (you get $100 + $200 on activity completion via the Free Tier). Either is fine.
• Read through the quickstart README and the linked tutorial at https://iii.dev/docs/quickstart end-to-end before you start provisioning anything. Understand what each worker does and how they call one another before you split them across machines.

What to build

1. Provision the network. Create a VPC with a private subnet in your chosen cloud. The VMs that host the workers must not be directly exposed to the public internet — only the API gateway VM should have a public-facing endpoint.
2. Deploy the workers across VMs. Run each worker from the quickstart project on its own VM inside the subnet. The workers must communicate with each other via RPC across the subnet (not co-located on a single box, and not over the public internet).
3. Expose inference as a JSON API. Stand up a front-door service (on its own VM or as a managed endpoint) that accepts HTTP requests with a JSON body, dispatches the request into the worker mesh, and returns the inference result as JSON. The request/response schema is your call — document it.
4. Make it reproducible. Anything you provisioned by clicking around in the console should also be expressible as code (Terraform, Pulumi, gcloud/aws CLI scripts, Ansible — pick one). We should be able to tear it down and bring it back up from your repo.

Deliverables

Submit a repository (public Git repo or a tarball) containing:

• Infrastructure-as-code for the VPC, subnet, VMs, and firewall rules.
• Deployment scripts or configuration (systemd units, container manifests, etc.) for each worker.
• A short README.md with:
  • An architecture diagram (ASCII or image) showing the subnet, VMs, and RPC flow.
  • The exact curl command that hits your JSON API, along with a sample request and response.
  • Instructions to redeploy the stack from scratch in a fresh cloud account.
• A short writeup (a few paragraphs is fine) covering: what you would harden before putting this in production, and what you would do differently if the model were 100x larger.

Evaluation criteria

• Correctness — the JSON API actually returns inference results end-to-end through the RPC chain.
• Network hygiene — workers are not reachable from the public internet; only the API endpoint is.
• Reproducibility — your IaC actually works on a clean account.
• Clarity — your README is enough for someone else on the team to redeploy and debug.

DO NOTE that even incomplete submissions are allowed provided we are able to see how you thought through the entire process. If you can document it, there are brownie points for it.

Timeline

Implementation and deployment should take typically 36-72 hours (2-3 days). Final Deadline: 23th May, 2026

Submission

Email your repo link (or tarball) to anuran@getalchemystai.com with the subject line DevOps Internship Assignment — <Your Name>. Add saumitra@getalchemystai.com and khushi@getalchemystai.com to CC as well.