Nuclear AI Tools of Awesomeness

Sonnet recommended · Haiku: for learning and basic functional testing only

How Nuclear AI Tools of Awesomeness Works

This is Mat's personal grouping of independent AI tools, currently three of them, all running on one Claude API key. They mostly don't depend on each other — pick the one you need from the toggle above the query box, and switching between them keeps each one's own conversation intact (nothing is dropped, it's just out of view until you switch back). Each tool is described in its own section below.

1. General ADAMS Research

Uses the NRC ADAMS Public Search (APS) API, server-side.

For open-ended lookups — find a document, check what's on file for a plant, read an amendment, answer a question. Claude searches, summarizes what it found in plain language, and shows a linked table of results. It does not open documents on its own: it stops there and asks what you'd like it to read — with an approximate cost — so you decide before any tokens are spent reading.

2. Design-Basis Change Analysis via ADAMS

Also uses the ADAMS API — a heavier, structured workflow than General ADAMS Research above.

For structured gap analysis — what changed in a system's licensing basis over a date range. Claude follows a five-bucket methodology and stops at two checkpoints to get your approval before doing expensive work.

Plan & confirm
Claude resolves the plant to its docket number(s) and lays out a five-bucket search plan (license amendments, 50.59 reports, UFSAR, ISI/relief requests, generic/vendor issues). It stops and waits for you to reply "go" before searching.
Search & triage
Claude runs all five buckets against ADAMS, rates every result (★★★ HIGH / ★★ MEDIUM / ★ LOW / — NOISE), and shows the triage. It stops again for your approval — and gives a token cost estimate — before opening documents.
Read & report
After you approve, Claude reads each document and produces a plain-language Design-Basis Change Analysis report with linked citations. Save it as a PDF with the button that appears on the final report.

At any checkpoint you can reply with a correction instead of "go" — change the dockets, the date range, the ratings, or anything else — and Claude adjusts before continuing.

3. Risk Management Plans

Fully independent of the other two — does not touch ADAMS at all.

Drafts a Qualitative Risk Management Plan (QL-RMP) following ENERCON's QPM-003 procedure and QF-034 template. Describe the project and/or upload source documents (.docx or .pdf); Claude identifies risk exposures, rates Likelihood and Impact, and produces a risk exposure table. The app computes each risk's color (Green/Yellow/Red) itself from a fixed matrix — Claude never picks the color directly. If your first message doesn't give Claude enough to work with, it will ask for more before drafting anything. The final draft downloads as a real Word document with the button that appears on it — always a first-pass draft meant for ENERCON's normal RMP review and approval process, not a signed, approved plan.

Data & privacy

Model choice

Sonnet 5 is the default and handles most searches well. Haiku is for learning the tool and basic functional testing only. Opus and Fable aren't offered here — left out on purpose to keep costs down.