Skip to content

Latest commit

 

History

History
30 lines (20 loc) · 1.58 KB

ASSESSMENTS_AI.md

File metadata and controls

30 lines (20 loc) · 1.58 KB

Assessments and grading (demo)

Assessments — create, take, submit, and grade — are implemented in the demo (Go app), not in the docproc CLI. docproc is document processing only.

Where it lives

  • Backend: demo/go/ — API routes for assessments, questions, submissions; grading in internal/grade/ (single-select, formula, conceptual, derivation).
  • Frontend: demo/web/ — Create assessments, take them, view results.
  • Database: PostgreSQL (from demo/docker-compose.yml): docproc_assessments, docproc_questions, docproc_submissions.

See demo/README.md for how to run the demo.

Grading modes (Go)

Type Mode Behavior
single_select / mcq single_select Deterministic: student choice matches correct choice → full marks, else 0.
short_answer (equation-like ref) formula LLM: "Are these two mathematical expressions equivalent?" → score 0–100.
short_answer (default) conceptual LLM: grade by coverage of key concepts; paraphrases get credit.
long_answer derivation LLM: rubric-based partial credit per step.

Grading uses the demo’s OpenAI client (OPENAI_API_KEY). No docproc CLI involved.

Submission flow

  1. POST /assessments/:id/submit with answers: { question_id: value, ... }.
  2. Go loads questions, infers mode per question, runs the grader (single_select in-process; formula/conceptual/derivation via LLM).
  3. Results are stored in docproc_submissions (score_pct, question_results) and returned.

No background queue for grading; submission is synchronous.