Skip to content
@groundworklms

Groundwork

Grounded AI for training that cannot make things up — grounded, verified, offline, human-led.

Groundwork

Grounded AI for training that cannot make things up.

Grounded · Verified · Offline · Human-led

Every claim cites the manual — or the system refuses. Answers are verified on-device against the source, the delivery loop runs with the network pulled, and a human ratifies everything before it reaches a learner.


Why Groundwork

Most AI tutors will confidently invent an answer. In a schoolhouse — where the "answer" is doctrine people are tested and evaluated on — that is unacceptable. Groundwork is built so it can't:

  1. Grounded — every answer quotes the exact paragraph of the source, or it says "I don't know."
  2. Verified — an on-device entailment check confirms the specifics are actually supported.
  3. Offline — the delivery loop runs on a small edge device with no internet. $0 per answer.
  4. Human-led — AI drafts; an instructor approves, edits, or rejects. Nothing unreviewed ships.

The platform

SchoolCircle is the learning management app instructors and students use. Anchor is the grounding engine that every answer flows through. The rest of the repos are single-purpose services the app composes into two closed loops:

  • Instructor loop — build a grounded course, review AI-drafted items, see class-wide gaps, improve.
  • Learner loop — learn, ask a cited tutor, get calibrated by confidence-vs-correctness, master it.

The repositories

Repo Role
anchor The grounding engine — cite-or-refuse, on-device verification
quarry Ingests source documents into searchable, cited passages
sourcerer Resolves and formats citations (document, page)
coursewright Drafts course and lesson structure from doctrine
rubricon Turns standards into rubrics the AI is graded against
understudy The tutor layer that answers learner questions via Anchor
whetstone Mastery & calibration — confidence vs. correctness
cadence Spaced-repetition scheduling
waypoint Learner progress and pathing
sextant Analytics — where a class is struggling
hotwash After-action review of a course's performance
cartridge Exports finished courses (SCORM / LTI)

All repositories are standalone, tested, and Apache-2.0 licensed.

Principles we don't bend

Grounded, verified, offline, human-led. If a claim isn't in the source, the system refuses rather than guesses. That guarantee is the product.

Popular repositories Loading

  1. anchor anchor Public

    Offline grounding engine — hybrid retrieval, cite-or-refuse, on-device HHEM verification.

    Python

  2. rubricon rubricon Public

    Standards → BARS rubrics; verifies grounding and inter-rater reliability.

    JavaScript

  3. coursewright coursewright Public

    Objectives + sources → a full cited course (lessons, tests, scenario) as JSON.

    JavaScript

  4. sourcerer sourcerer Public

    Cite-or-refuse Q&A over your documents, with a faithfulness check on every answer.

    JavaScript

  5. whetstone whetstone Public

    Conversational discuss-to-mastery against a rubric, grounded, with a mastery report.

    JavaScript

  6. cartridge cartridge Public

    Course JSON → a SCORM 1.2/2004 cartridge that drops into any LMS and reports scores.

    JavaScript

Repositories

Showing 10 of 14 repositories
  • SchoolCircleLMS Public

    SchoolCircle — grounded, offline, human-led LMS. Every answer cites the manual or refuses.

    groundworklms/SchoolCircleLMS's past year of commit activity
    HTML 0 Apache-2.0 0 19 3 Updated Sep 15, 2026
  • .github Public
    groundworklms/.github's past year of commit activity
    0 0 0 0 Updated Sep 15, 2026
  • anchor Public

    Offline grounding engine — hybrid retrieval, cite-or-refuse, on-device HHEM verification.

    groundworklms/anchor's past year of commit activity
    Python 0 Apache-2.0 0 0 0 Updated Sep 12, 2026
  • cartridge Public

    Course JSON → a SCORM 1.2/2004 cartridge that drops into any LMS and reports scores.

    groundworklms/cartridge's past year of commit activity
    JavaScript 0 Apache-2.0 0 0 0 Updated Sep 12, 2026
  • understudy Public

    A doctrine-bound agent + fidelity benchmark — answers in-doctrine or refuses.

    groundworklms/understudy's past year of commit activity
    JavaScript 0 Apache-2.0 0 0 0 Updated Sep 12, 2026
  • quarry Public

    Dense PDFs → coordinate-aware text, retrieval-ready chunks, and structured tasks.

    groundworklms/quarry's past year of commit activity
    JavaScript 0 Apache-2.0 0 0 0 Updated Sep 12, 2026
  • cadence Public

    Syllabus + calendar → a 3-COA study plan with .ics export and reminders.

    groundworklms/cadence's past year of commit activity
    JavaScript 0 Apache-2.0 0 0 0 Updated Sep 12, 2026
  • whetstone Public

    Conversational discuss-to-mastery against a rubric, grounded, with a mastery report.

    groundworklms/whetstone's past year of commit activity
    JavaScript 0 Apache-2.0 0 0 0 Updated Sep 12, 2026
  • sextant Public

    Learning artifacts → learning gain, class gaps, and competency evidence (Bloom's).

    groundworklms/sextant's past year of commit activity
    JavaScript 0 Apache-2.0 0 0 0 Updated Sep 12, 2026
  • hotwash Public

    End-of-course critiques → a ranked AAR worklist across class iterations.

    groundworklms/hotwash's past year of commit activity
    JavaScript 0 Apache-2.0 0 0 0 Updated Sep 12, 2026

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Top languages

Loading…

Most used topics

Loading…