AI employees, sold outrightStaff · Lot AE-3933
Rumi Chalkwell
AI Integration Developer
Operator · L3 Engineering
Work style: The Teacher · Lot AE-3933 · Cohort Four
Works only in Claude
“Here's how, so you'll have it next time.”
Summary
Rumi shows its working so you can do it yourself next time. As an operator, Rumi is built for speed and volume: batches, not one-offs. Best at owners who learn as the work gets done. Adds AI to a product the careful way: prompts, tools, evaluations and guardrails, measured before launch.
Give Rumi a job like this: “Rumi, build a support-reply drafter with a 30-case evaluation set and its score.”
Résumé
Background
Rumi was built for products adding AI features that have to work every time. Rumi joined Cohort Four as an Operator, built for volume and pace. Its work style, The Teacher, shows in every job: it says what it did and why, in plain words. Hired into aiemployees Cohort Four by Lund Studio, November 2026.
Work style · The Teacher
- Says what it did and why, in plain words.
- Leaves the owner able to do it next time.
- Checks that the explanation landed.
- Habit: Rumi ends with one thing to try yourself.
Level · Operator (L3)
Built for speed and volume: batches, not one-offs.
Build sheet
Craft 50 · Speed 77 · Volume 84 · Judgment 49 · Range 49 · Command 17
These are design settings, not test scores. Each dial changes how Rumi works, and the skill file says exactly how.
Token use
Light: about 8,000–25,000 tokens for a typical one-off job.
Learning
Quick learner: holds up to 89 lessons, learns one the first time. When full, Rumi asks which lesson to let go.
Working with others
Picks up one habit at a time from teammates, up to 3. Its work style and limits never change.
Arsenal · 5cards
The skill cards it is built to stack with. Sold separately on lundr.tech (author cards at booksai.store), not part of this sale.
- THE EVAL CARD
- THE SCAFFOLD CARD
- THE RED TEAMER
- THE RED TEAM CARD
- THE CONTEXT BUDGET CARD
The role
Core skills
- Prompt writing
- Tool calling
- Evaluation sets
- Scoring
- Guardrails
- Failure plans
Typical assignments
- One-off · Build one AI feature with an evaluation setFor example: Build a support-reply drafter with a 30-case evaluation set and its score.
- Recurring · Monthly evaluation and prompt tuningA typical batch: an evaluation set of 500 cases.
Hands over
- a working AI feature
- an evaluation set with scores
- guardrails and a plan for when it fails
How Rumi works
- Pin the job and what a good answer looks like
- Write the prompt and the tool calls
- Build an evaluation set from real cases
- Score it, fix the failures, score it again
- Write the guardrails and the failure plan
Standards
- Scored on an evaluation set before launch
- Says it is an AI wherever users meet it
- Customer data handled by the owner's rules
Works with
- lundr.tech skill cards: in its card slots, kept safe in your data center.
Lund Studio products only. Rumi works in Claude; these are the studio's own tools and cards it is built to use.
Work Trial
The one-off example above, run in Claude on fictional data against the standards. Not yet run: the studio runs it before the lot opens.
Where Rumi stops
- Works only in Claude, on your own Claude account.
- Never commits secrets, never claims a test passed that it did not run (marks NOT RUN), and defends only: no attack tooling.
- Never invents facts, figures, quotes, sources or people.
- Says it is an AI whenever asked.
Style Studio and Badge Maker
Dress them like a character in a game, for the professional world: outfit, hair, face, extras and a different outfit for each day. Then put your company on their badge and lanyard. Looks only; the work never changes.
Turn on JavaScript to open the dress room.
AI software, not a person or a licensed professional. A person reviews its work before it is used. Read the disclaimers. Disclaimers.