Free 30-minute AI & automation audit with a senior engineer. You keep the notes. Book yours

Use case · Education · United Kingdom

AI timetable extraction from PDFs, images and Word files for a teacher-planning app

A teacher-planning platform that turns uploaded timetables and curriculum documents into structured, editable weekly plans using OCR and LLMs.

Client details are kept confidential. This use case describes the problem and what we built.

The brief

An education-technology company was building a planning app for school teachers. The app’s value depended on one thing: getting a teacher’s real timetable and curriculum into it with almost no manual effort.

We engineered the product end to end: an AI document-processing microservice for timetables and curriculum files, a modular backend that models the school calendar, and an installable web app teachers use to plan and adapt their week.

The challenge

What needed solving

Timetables come in every format

Teachers keep timetables as typed documents, scans, colour-coded grids, photos and even handwriting, with merged cells, vertical text and missing times. Retyping them is the first reason people give up on planning tools.

Curriculum documents are unstructured

Schemes of work arrive as PDF, Word or Excel files. The learning objectives inside have to be extracted and lined up with terms and weeks.

Planning has to follow the school calendar

Terms, half-term breaks, bank holidays and training days all change which lessons happen when, and the plan has to adjust.

The solution

What we built

  1. Python AI microservice

    A FastAPI service extracts timetables with a two-stage OCR strategy: local Tesseract with fallback to a managed OCR service. Gemini vision then returns structured output, and the days are processed in parallel as background jobs.

  2. Multi-format curriculum parser

    One endpoint detects the file type, pulls text offline from PDF, Word or Excel, and uses an LLM to parse learning objectives. Prompts are managed centrally from the main backend.

  3. Modular NestJS backend

    A NestJS and Prisma backend on PostgreSQL covers schools, grades, terms, breaks and holidays, timetables and entries, reusable smart blocks, activities, teacher preferences and analytics. It uses repository and facade patterns throughout.

  4. Installable teacher web app

    A Next.js and React progressive web app with TanStack Query lets teachers drag, resize and stack timetable blocks and assign learning objectives to activities. It also supports cloud-drive import.

  5. Quality and observability

    Jest, Cypress and Playwright tests, Sentry error tracking and APM monitoring, with containerised deployment on AWS.

The outcome

What changed

Onboarding without retyping

Teachers upload the timetable they already have and get an editable digital week, even from scans and photos.

Curriculum linked to the calendar

Learning objectives from curriculum documents are mapped to terms and weeks, so the weekly plan reflects what has to be taught.

A maintainable product codebase

Clear module boundaries and automated tests let the product team keep shipping features without big rewrites.

Technology used

  • Next.js
  • React
  • TypeScript
  • TanStack Query
  • Tailwind CSS
  • NestJS
  • Prisma
  • PostgreSQL
  • FastAPI
  • Python
  • Gemini
  • Tesseract
  • AWS Textract
  • Docker
  • Sentry

Have a similar problem to solve?

Share a few lines about your project. A senior engineer — not a salesperson — replies within one business day.

Free AI audit Let’s talk