Custom AI Grading Software for Dental Schools

Evan Menke, a student at the University of Colorado, had already proven the hard part. Using AI in a personal capacity, he was evaluating dental crown preparations, the physical reproductions dental students carve during training. A student finishes a crown, photographs it, runs it past the AI, and gets a numeric score against a rubric plus an explanation of where it fell short. Faculty time saved, and students self-correct between formal reviews. What he did not have was AI grading software: a real platform that could serve students across multiple dental schools rather than one person's workflow.

Getting there meant moving from a single-user experiment to a backend that could handle multiple rubrics, switch between LLM providers as they evolve, store scoring history per student, and integrate with a frontend faculty and students could actually use. No off-the-shelf grading tool covers that combination, especially for photographed physical work rather than essays or quizzes.

Evan brought Leanware in to build the evaluation engine, the rubric system, and the multi-model orchestration that would keep the platform accurate as the underlying LLM ecosystem moved. The frontend already lived on Base44; the work was everything behind it.

The engagement was scoped as a managed custom AI agent: Leanware builds, deploys, and operates the system; the client uses it. Fixed scope, one senior full-stack engineer, with the option to extend into managed-service operation once the platform reaches steady state.

An evaluation engine built to outlive any single model

The backend runs on FastAPI with LangGraph orchestrating the LLM workflows and PostgreSQL holding rubrics, evaluation history, and scoring metadata. The provider layer abstracts across OpenRouter, OpenAI direct, Google, and Anthropic, so the platform can swap models per rubric or per evaluation and run cross-provider comparisons without code changes. When a better model ships, the grading software adopts it instead of being stranded on last year's choice.

Rubrics the client controls

The configuration surface is the part of AI grading software that buyers tend to underestimate. Whoever owns the rubric can configure the base prompt the LLM sees, the rubric structure (which criteria count, and in what proportion), the score weights per criterion, and the thresholds at which the system flags a critical finding versus a minor one. Dental rubrics evolve as faculty refine their teaching, and the same backend handles each new shape without a rebuild.

Grading that shows its work

For each evaluated image, the system returns percentage-based coordinate highlights for the regions the LLM scored against. The frontend overlays the model's reasoning directly on the student's photo, turning an abstract score into a specific "here is where the margin is wrong" annotation.

Tested before it graded anyone

Deployment runs from GitHub Actions to Render.com via Docker images. Before launch, Leanware ran accuracy testing across providers and delivered a comparative report covering which models scored highest on rubric adherence, which produced the most stable outputs across image variations, and which combinations were cost-effective at scale. Documentation and code shipped at a standard that supports ongoing maintenance and a future expansion into training a custom model specialized for dental rubrics.

The backend launched against the Base44 frontend and is in active use evaluating crown preparations. Multi-provider switching works in production, and the rubric and prompt configuration surface lets Evan refine the grading system as more student work flows through it. The model comparison report stands as a published reference for which provider performs best on which rubric type.

The collaboration continues on a smaller cadence. As evaluation samples accumulate, the path opens to training a custom model specialized for dental crown rubrics, which would lift accuracy beyond what general-purpose LLMs produce today.