AI-Powered Repository Evaluation Assistant

Grade student repositories in minutes, not hours.

RubricLM connects to student GitHub repositories, drafts weighted rubrics grounded in real code evidence, and delivers publication-ready Excel workbooks — with every score cited, flagged, and fully overrideable.

CS-301 · Project 2 — REST API

42 of 44 graded · rubric v2 · run #7

● Live
Ana Petrova
8979
33/40
Marcus Lee
7!88
29/40
Priya Nair
998
35/40

Amber = flagged for review · Purple = instructor override · every cell links to cited code

15 min → 90 sec

Per-student grading time

80%+

AI scores accepted unchanged

300 × 8

Gradebook cells at full frame rate

5 sheets

In every exported workbook

Workflow

From assignment brief to final workbook

Five steps from repo links to registrar-ready grades — and you approve everything along the way.

  1. 01

    Connect & import

    Sign in with read-only GitHub access, create a course, then upload a CSV or paste repo URLs. Every link is validated and branch-resolved first.

  2. 02

    Draft the rubric

    RubricLM proposes weighted criteria with performance anchors from your assignment brief. Tune weights until they sum to exactly 100%.

  3. 03

    Run the evaluation

    Repos are cloned and analyzed statically, then scored criterion by criterion while a live board streams progress end to end.

  4. 04

    Review with evidence

    Open any score for cited files, highlighted line ranges, and AI commentary. Override in one click — original scores are preserved.

  5. 05

    Export the workbook

    Download a formatted XLSX with Summary, Rubric Reference, per-student Detail sheets, Evidence Log, and Raw Metrics.

Features

Built for rigor, designed for speed

Automation where it saves time, human judgment where it matters.

🧾

Evidence-grounded scoring

Every score carries at least one citation: file path, line range, and a verbatim quote verified against the repository snapshot.

⚖️

Rubrics in minutes

Four to eight weighted criteria with scoring anchors, drafted from your brief, then editable, reorderable, and fully versioned.

🔬

Static analysis only

LOC, language breakdown, test detection, commit stats, README quality, CI presence — computed without ever executing student code.

🚩

Confidence flags & overrides

Low-confidence scores arrive pre-flagged. Overrides keep the original AI score, author, and timestamp for later calibration.

📡

Live run monitoring

Watch every submission move through clone, extract, grade, and aggregate stages on a streaming status board with smart retries.

📗

Registrar-ready exports

Multi-sheet Excel with frozen headers, color scales, and clickable GitHub permalinks — verified across Excel, Sheets, and LibreOffice.

Trust

Defensible by design

Automated grading only works if you can defend it to a department chair. RubricLM pairs tenant isolation and encryption in transit and at rest with transparent evidence, effortless overrides, and a complete audit history of every action taken in your workspace.

Read-only by design

We request minimum viable GitHub scopes — never write or admin.

FERPA-minded posture

Workspace-scoped records with a documented retention window and automated purge.

Secret scrubbing

Keys, tokens, and connection strings are stripped before any excerpt is stored.

Immutable audit trail

Every override, finalize, and export is logged with actor and timestamp.

Reclaim your weekends this term.

RubricLM is onboarding a limited cohort of pilot instructors for the upcoming semester. Bring one course — we will take it from there.

Free during beta · No credit card · Works with any GitHub-hosted course