Business Analytics Tutoring Assistant

The Problem

MBA students learning optimization modeling hit a wall outside of office hours. The mechanics are unforgiving — SUMPRODUCT layouts, Big-M linking constraints, Solver configuration — and textbooks don't walk through a specific problem with you at 11 p.m. on a Sunday.

The Approach

Built a browser-based tutor that takes a problem in almost any format — PDF, Word, Excel, or plain text — and returns a structured six-step walkthrough plus a ready-to-run Excel workbook with live SUMPRODUCT formulas and a Solver setup instruction sheet.

Underneath, Claude is forced to return a strict JSON schema (via Anthropic tool use) covering problem type, decision variables, objective, grouped constraints, Excel layout, and Solver dialog settings — with required why / purpose / explanation fields so the model has to justify its reasoning rather than hand-wave answers.

The Outcome

A single static HTML file, deployable anywhere. No backend to run, no database to secure, no per-student billing to manage. Students bring their own Anthropic API key; their problem text and API key never leave their browser. Built to be copied, self-hosted, or forked — pedagogy as open infrastructure.

Stack

• Anthropic Claude API (direct from browser)
• Claude tool use (structured JSON output)
• Model picker: Sonnet 4, Sonnet 4.5, Opus 4.5, Haiku 4.5
• Native PDF & image input (no OCR)
• mammoth.js (.docx parsing)
• SheetJS (.xlsx parsing & generation)
• Retry with exponential backoff
• Zero backend · static HTML · CDN-hosted deps

Privacy Posture

API key stored only in browser localStorage. Problem text and uploaded files go directly to Anthropic's API for inference — nowhere else. No analytics, no telemetry, no third-party calls beyond the Google Fonts, mammoth.js, SheetJS CDN loads, and the Anthropic API itself.