About Chemistry Tutor

1Software Information

Version: 0.6.49

Date: July 24, 2026

Developers: Jeff Osborne and Cursor AI (Claude Opus, Grok 4.5)

Location: Manchester University, North Manchester, Indiana, USA

Hosted by: Osborne Lab

2Algorithm & Design

Tutor, not autograder: Feedback is the deliverable. There is no score; revealing a worked answer is always an explicit, opt-in choice.

Deterministic chemistry engine: Every product-structure decision is made by a rule-based RDKit detector library. The same drawing always earns the same feedback.

No LLM in the chemistry path: Large language models are never used to decide what is chemically right or wrong. Judgments stay reproducible and grounded in cheminformatics.

Misconception hits: Detectors recognize common errors (wrong regiochemistry, stereochemistry, directing effects, oxidation level, and many reaction-class traps) and return severity-coded explanations of the underlying concept.

Answer-leakage protection: Network responses withhold expected structures until an accepted attempt or an explicit solution request, so solving stays honest.

Parallel pathways: Structure drawing (Ketcher), curved-arrow mechanisms, ranking (e.g. acidity), multiple-choice spectroscopy/naming/biochemistry items, and metabolic-flux simulations share one catalog and UI.

3Key Features

Draw the product: Reaction problems with a browser chemical editor; submit a SMILES-backed structure and get targeted misconception feedback.

Curved-arrow mechanisms: Tap source and sink anchors (or drag source to sink) to draw electron-flow arrows; a verification gate checks each step, with pattern-level feedback when arrows miss.

Ranking & choice: Order molecules on an author axis (acidity, stability) or answer naming / IR / NMR / MS vignettes.

Graduated reveal: Condition notes, reaction type, mechanism summary, and full reference material unlock step by step so you can do as much as you can first.

Course learning objectives: CHEM 311 / CHEM 312 / CHEM 405 maps link practice problems to syllabus outcomes.

Related practice: Structure and mechanism problems for the same named reaction are co-located in the list and cross-linked on the detail page.

Optional accounts: Fully usable signed out. Pseudonymous username + password (no email) saves attempt history to your account when you sign in.

Instructor authoring: Seed sync, problem editing, and attempt analytics for instructors (API key or instructor account). For free-play curved-arrow checks against any golden (without a catalog problem), use the mechanism sandbox.

Privacy: Tutoring runs on this server; accounts store no email or other PII beyond a chosen username.

4Technical Stack

Backend: Python 3.9+ with FastAPI and Pydantic

Chemistry engine: RDKit-backed misconception_library detectors plus a separate curved-arrow mechanism gate

Concept registry: YAML reaction reference library (mechanism summaries, teaching notes, credited diagrams)

Structure editor: Ketcher in an iframe

Frontend: HTML5, CSS3, vanilla JavaScript (ES2020)

Storage: SQLAlchemy Core over SQLite (default) or Postgres; opt-in attempt logging

Server: uvicorn on loopback, with Apache reverse proxy at /chemtutor

5Contact & Support

For questions, bug reports, or feature requests, please contact Jeff Osborne.

Email: jposborne@manchester.edu

Return to the problem list to get started, or browse course objectives. Instructors can open the authoring UI, the mechanism sandbox for free-play golden checks, or arrow polish to nudge display bends and tip placement on on-disk goldens.