Query Letter Copilot

Paste your query letter, get instant AI feedback on hook, clarity, and genre fit — plus a stronger rewritten opening. Powered by NVIDIA Nemotron via Nebius Token Factory.

Comment

Inspiration As a freelance writer, I've pitched dozens of personal essays and reported features to outlets like Vox First Person, Business Insider, and Noema. Every time, I've wondered: is my opening paragraph actually strong enough to hook an editor, or am I about to hit send on something forgettable? I wanted a fast, honest second opinion — before the pitch goes out, not after a rejection comes back.

What it does Query Letter Copilot lets you paste your query or pitch letter, choose a genre, and get instant feedback across four dimensions: Verdict (an overall read on the pitch), Hook strength (does the opening grab attention?), Clarity (is the ask easy to follow?), and Genre fit (does it follow the conventions editors expect for that genre?). It also rewrites your opening paragraph, giving you a stronger version to work from.

How we built it The backend is a small Node.js + Express server that sends the query letter and genre to NVIDIA's Nemotron model (NVIDIA-Nemotron-3-Nano-30B-A3B) via Nebius Token Factory, using a structured prompt that asks for JSON output. The frontend is plain HTML/CSS/JavaScript — no framework, just a clean form and result cards. The whole thing is deployed on Google Cloud Run, built directly from source with Cloud Build.

Challenges we ran into Getting from a working local prototype to a public deployment was the hardest part. Cloud Run's build process needed several IAM permissions (Artifact Registry read/write, Cloud Logging write) that weren't obvious from the initial error messages — each one only surfaced after fixing the last. I also worked through GitHub's new mandatory 2FA rollout and some local environment hiccups along the way. None of it was glamorous, but each fix taught me something about how Cloud Run's build pipeline actually works under the hood.

Accomplishments that we're proud of Going from no Nebius account to a fully deployed, publicly accessible AI tool in one focused session — with a real backend, a working frontend, and a clean public GitHub repo — felt like a genuine accomplishment. It's also a deliberate change of pace from my usual medical and regulatory projects: something built for my own field, freelance writing.

What we learned Model access is the easy part — Nemotron via Nebius Token Factory worked correctly on the first real test. The friction lives in the deployment plumbing: service account permissions, build logs, and reading errors carefully enough to fix one root cause at a time instead of guessing.

What's next for Query Letter Copilot Letting users save and compare multiple drafts of the same pitch, adding outlet-specific guidance on what different publications tend to look for, and supporting a combined pitch-plus-manuscript read for a fuller picture before submission.

Built With

Share this project:

Updates