Autograding for Gradescope — turn your existing lab manual into a structured, AI-graded Gradescope assignment in under an hour, or run timed multiple-choice reading checks straight from a SQLite question bank.
GradeBridge AI is a privacy-first system with two complementary pipelines:
- Lab Pipeline — for engineering lab reports, mini-projects, and homework with mixed text, image, and AI-graded responses. Drop in a lab manual; ship to students; receive autograded results in Gradescope.
- MQ Pipeline (new, May 2026) — for timed multiple-choice reading checks and quizzes drawn from a portable SQLite question bank. Useful when an LMS quiz tool is unavailable or you want vendor-independent question delivery.
Both pipelines share the same AES-256-GCM encryption and the same Gradescope-Docker autograder pattern. No student data passes through any VeriQAi server at any point.
- Read the Instructor Overview — a plain-English walkthrough of the full pipeline (10 min read)
- Browse the worked example — a fictional lab assignment taken from source document all the way through to the TA grader document
- When ready, follow the Step-by-Step Usage Guide to build your first assignment
%%{init: {'theme': 'default', 'themeVariables': {'fontSize': '18px', 'fontFamily': 'arial'}}}%%
flowchart TD
A["📄 Your lab manual or project description"]
B["🤖 Claude Code\nGenerates structured assignment file"]
C["⚙️ Assignment Maker\nReview · adjust · export"]
D["📦 Student package\nDistributed via Canvas"]
E["📋 Grader document\nConfidential TA reference"]
F["✏️ Student Submission App\nStudents fill in answers in browser"]
G["🎓 Gradescope\nAI autograding and TA review"]
A --> B --> C
C --> D
C --> E
D --> F --> G
style A fill:#f5f5f5,stroke:#9e9e9e,color:#1a1a1a
style B fill:#e8f4fd,stroke:#1565c0,color:#1a1a1a
style C fill:#e8f4fd,stroke:#1565c0,color:#1a1a1a
style D fill:#e8f5e9,stroke:#2e7d32,color:#1a1a1a
style E fill:#fff8e1,stroke:#f57f17,color:#1a1a1a
style F fill:#e8f5e9,stroke:#2e7d32,color:#1a1a1a
style G fill:#fce4ec,stroke:#c62828,color:#1a1a1a
What each step involves:
- Source document — your existing Word doc, PDF, or Markdown lab manual. No reformatting required.
- Claude Code — reads the source and proposes question types, point values, and grading rubrics. You confirm or correct before rubrics are written.
- Assignment Maker — browser app (no install). Import the generated file, review every question, adjust anything, export.
- Student Submission App — browser app (no install, no account). Students load the assignment, fill in answers, and click one button to download their Gradescope submission.
- Gradescope — receives a standard ZIP upload. AI-graded questions are scored automatically; human-reviewed questions are flagged for TA attention with the grader document as reference.
%%{init: {'theme': 'default', 'themeVariables': {'fontSize': '18px', 'fontFamily': 'arial'}}}%%
flowchart TD
A1["💾 SQLite question bank\n(modules → chapters → topics → questions)"]
B1["⚙️ MQ Assignment Maker\nFilter, pick pool, set time/N, export"]
C1["🔐 Encrypted assignment .json\n(+ optional 1.5x / 2x accommodation variants)"]
D1["⏱️ MQ Student Submission App\nTimed quiz, honor pledge, encrypted ZIP"]
E1["🎓 Gradescope MQ Autograder\n(Python, Docker, identity check)"]
A1 --> B1 --> C1 --> D1 --> E1
style A1 fill:#fff8e1,stroke:#f57f17,color:#1a1a1a
style B1 fill:#e8f4fd,stroke:#1565c0,color:#1a1a1a
style C1 fill:#f3e5f5,stroke:#6a1b9a,color:#1a1a1a
style D1 fill:#e8f5e9,stroke:#2e7d32,color:#1a1a1a
style E1 fill:#fce4ec,stroke:#c62828,color:#1a1a1a
What each step involves:
- Question bank — your SQLite database of multiple-choice questions, organized by module, chapter, and topic. You own the file; nothing leaves your machine.
- MQ Assignment Maker — browser app (no install). Drop in the
.dbfile, pick a pool, set time limit and N-per-student, export an encrypted assignment file. One click generates accommodation variants (1.25x, 1.5x, 2x, or custom) for SDC students. - Encrypted assignment file — distributed to students. Cannot be read or edited in a text editor; tampering is detected automatically.
- MQ Student Submission App — browser app. Student enters name, loads the assignment, acknowledges Honor Code expectations, takes a timed quiz one question at a time (with palette navigation), signs the pledge, downloads an encrypted ZIP for Gradescope.
- Gradescope MQ Autograder — Python autograder running in Gradescope's Docker. Decrypts submission, scores against the encrypted spec baked into the image, returns Gradescope
results.jsonwith per-question feedback. Compares the typed student name to the Gradescope-authenticated submitter and flags mismatches for instructor review.
| App | Link | |
|---|---|---|
| Lab pipeline | ||
| For instructors | Assignment Maker | Open app |
| For students | Student Submission | Open app |
| MQ pipeline | ||
| For instructors | MQ Assignment Maker | Open app |
| For students | MQ Student Submission | Open app |
The examples/ folder contains a complete end-to-end example using a fictional engineering course:
| File | Description |
|---|---|
source_lab_manual.md |
The instructor's starting point — a typical lab manual as it exists before GradeBridge AI |
assignment.md |
The structured assignment file generated by Claude Code from the source |
grader_document.html |
The TA grader reference — open in any browser; LaTeX renders automatically |
Security note: Student submission files are encrypted with AES-256-GCM before download. The encrypted submission can only be read by the Gradescope autograder using the private grading rubric produced at export time.
GradeBridge AI supports six question types. Claude Code classifies each question automatically — instructors review and adjust.
| Type | Student does | Graded by |
|---|---|---|
| Text | Types a calculation or short answer | TA |
| Image | Uploads a photo or screenshot | TA (checks against grader checklist) |
| AI Graded: Binary | Writes a yes/no answer with brief justification (~20 words) | AI autograder |
| AI Graded: Short | Writes a focused answer on one concept (~50 words) | AI autograder |
| AI Graded: Medium | Explains a mechanism or comparison (~100 words) | AI autograder |
| AI Graded: Long | Analyses trade-offs or evaluates a design (~150 words) | AI autograder |
| Document | What it covers |
|---|---|
| Instructor Overview | Narrative introduction to the full pipeline — start here |
| Step-by-Step Usage Guide | How to run Claude Code, import, review, and export |
| Worked Example | Complete example: source → assignment file → grader document |
| Document | What it covers |
|---|---|
| Assignment File Format | Full specification for the .md assignment file format (lab pipeline) |
| Claude Code Prompt | Ready-to-use prompt for generating an assignment from a lab manual |
| Python Converter | Local alternative to browser import: converts .md to assignment_spec.json |
| Submission Encoding Spec | AES-256-GCM encoding specification (shared by lab and MQ pipelines) |
| Autograder ZIP Spec | Gradescope lab autograder integration and ZIP extraction spec |
| MQ Autograder Brief | Implementation spec for the MQ Gradescope Docker autograder |
| Traditional Workflow Spec | PDF-only Gradescope workflow (no AI autograding) |
GradeBridge-AI/
├── examples/ ← worked example: source → assignment → grader doc
├── CCAssignmentMaker/ ← Lab pipeline: CC prompt, format spec, Python converter, autograder
├── GradeBridge-MQ-Autograder/ ← MQ pipeline: Python autograder for Gradescope Docker
├── GradeBridge_Instructor_Overview.md
├── AUTOGRADER_ZIP_SPEC.md
├── MQ_AUTOGRADER_BRIEF_2026-05-08.md ← spec for the MQ Docker autograder implementer
└── TRADITIONAL_WORKFLOW_SPEC.md
The four browser apps live in separate repositories:
Lab pipeline:
MQ pipeline:
MIT License. Copyright (c) 2026 VeriQAi. See LICENSE.