Schedule Upload: Parsing UTM SGU PDFs with Inforario

Inforario accepts your official UTM SGU schedule PDF and transforms it into a structured, interactive timetable without any manual data entry. Drop the file onto the upload zone, and Inforario will automatically detect subjects, teachers, time slots, and room codes — then resolve any overlapping sessions before handing the data to the viewer.

The DropZone Component

The DropZone component is the entry point for all uploads. It accepts drag-and-drop events as well as a standard file picker, validates the file type, and delegates to the useScheduleUpload hook for processing.

Select or drag your PDF

Click Seleccionar Archivo or drag your SGU schedule PDF directly onto the upload zone. The component listens for dragenter, dragover, dragleave, and drop events and provides visual feedback while a file is being dragged over it.

File validation

Only application/pdf files are accepted. If you drop an unsupported file type, an inline error banner appears with an × dismiss button. Image files are processed through the OCR fallback path in the underlying hook but are not shown in the main drop zone UI.

Confirm and process

Once a valid file is selected, its name and size are displayed. Click Procesar Horario to start parsing. The button shows a spinning loader and the text Analizando Documento… while the file is being processed.

Success handoff

On success, useScheduleUpload calls the onSuccess(schedule) callback, which transitions the app from AppView.LANDING to AppView.DASHBOARD with the parsed Schedule object.

Parsing Pipeline

useScheduleUpload orchestrates a two-path pipeline for PDFs and a separate OCR path for images. For PDFs, it first attempts AI-assisted extraction; on failure it falls back to the local regex parser.

AI Extraction (Primary)
Local Regex Parser (Fallback)
OCR Fallback (Images)

The AI path calls parseScheduleFileWithEdge from supabaseClient.ts. This function first extracts plain text from the PDF using pdfjs-dist, groups text items by row (y-coordinate proximity ≤ 2 px), then sends the reconstructed plain text to the extract-schedule Supabase Edge Function, which calls Groq’s llama-3.3-70b-versatile model to return a structured session list.

// supabaseClient.ts — AI extraction path
export const parseScheduleFileWithEdge = async (base64Data: string) => {
  if (!isSupabaseConfigured()) {
    throw new Error('Supabase no está configurado para usar extracción por IA.');
  }

  const pdfText = await extractPdfText(base64Data); // pdfjs-dist → plain text

  const { data, error } = await supabase.functions.invoke('extract-schedule', {
    body: { pdfText },
  });

  if (error) throw new Error(error.message || 'No se pudo invocar extract-schedule.');

  const sessions = mapSessions(data);   // validate & normalize AI response
  if (!sessions.length) throw new Error('La IA no devolvió sesiones válidas.');

  return { sessions };
};

The mapSessions helper validates each session returned by the AI: it checks that startTime and endTime match HH:MM format, maps Spanish day names to DayOfWeek, strips TECNOLOGÍAS DE LA prefixes and itinerary suffixes from subject names, and drops any entry missing required fields.

If Supabase is not configured (missing VITE_SUPABASE_URL / VITE_SUPABASE_ANON_KEY), the AI path is skipped entirely and the local parser is used directly.

When the Edge Function is unavailable or throws, useScheduleUpload falls back to parseScheduleFile from sguRegexParser.ts. This parser uses pdfjs-dist to extract every text item with its exact x/y canvas coordinates, then uses fixed column boundaries to classify each item.

// sguRegexParser.ts — column boundary constants
const COL = {
  SUBJECT_MAX_X:  170,   // items with x < 170 → subject name column
  DOCENTE_MIN_X:  260,   // items with 260 ≤ x < 415 → teacher column
  DOCENTE_MAX_X:  415,
  HORARIO_MIN_X:  495,   // items with x ≥ 495 → schedule / room column
};

Items in the schedule column whose text matches the pattern LUNES|MARTES|MIÉRCOLES|JUEVES|VIERNES (HH:MM:SS-HH:MM:SS) are treated as entry headers. All other schedule-column items below a header are associated with that header to extract room codes and locations.Each subject block is associated with the nearest entry header by vertical distance, producing a complete ClassSession entry.

If an image/* MIME type is passed to parseScheduleFile, the parser imports Tesseract.js dynamically and runs Spanish OCR on the image data URL. Progress is logged to the console during recognition.

// sguRegexParser.ts — OCR via Tesseract.js
async function parseImage(base64Data: string, mimeType: string): Promise<ParseResult> {
  const Tesseract = await import('tesseract.js');
  const imageDataUrl = `data:${mimeType};base64,${base64Data}`;

  const result = await Tesseract.recognize(imageDataUrl, 'spa', {
    logger: (m) => {
      if (m.status === 'recognizing text') {
        console.log(`OCR Progress: ${Math.round(m.progress * 100)}%`);
      }
    }
  });

  return parseRawText(result.data.text); // re-uses line-based parser
}

OCR accuracy depends heavily on scan quality. Blurry or low-contrast images may produce incomplete or incorrect schedules. Always prefer a PDF export from the SGU portal when available.

Metadata Extraction

The local parser scans text items in the top portion of page 1 (y < 210) to find the schedule header fields. Each label is matched by text content and its value is read from the next item on the same horizontal line (|candidate.y − label.y| ≤ 5).

PERIODO

Academic period range — normalized from "ABRIL DE 2026 HASTA AGOSTO DE 2026" to "ABRIL 2026 - AGOSTO 2026".

FACULTAD

Faculty name, stored in uppercase and carried into every ClassSession as subject_faculty.

ESCUELA

Career / school name, mapped to the career field in ParseResult.

ESTUDIANTE

Student full name, exposed as student_name in ParseResult.

The NIVEL field is also extracted but used internally for filtering only.

Normalization Logic

Subject Names

Raw subject names from the PDF often contain UTM-specific suffixes and prefixes that are stripped during parsing:

sub.name = sub.items
  .sort((a, b) => a.y - b.y)
  .map(i => i.text)
  .join(' ')
  .replace(/\s*\([A-Z0-9\s-]+\)\s*/gi, '')        // removes mesh codes like (A19)
  .replace(/^(TECNOLOGÍAS DE LA\s*)+/i, '')        // strips repeated prefix
  .replace(/\s+/g, ' ')
  .trim();

Teacher Names

UTM stores teacher names in the format LASTNAME1 LASTNAME2 FIRSTNAME1 FIRSTNAME2. The parser rearranges them to Firstname Lastname and strips academic title prefixes:

// Titles stripped: Ing., Lic., Dr., Dra., MSc., Mgtr., PhD, Abg., Arq., Econ., Prof.
const TITLE_PREFIXES = /\b(ing|lic|dr|dra|msc|mgtr|phd|abg|arq|econ|prof|sr|sra|srta)\.?\s*/gi;

// UTM 4-part name → "Firstname Lastname"
if (parts.length >= 3) {
  const firstName = capitalize(parts[2]); // NOMBRE1
  const lastName  = capitalize(parts[0]); // APELLIDO1
  return `${firstName} ${lastName}`;
}

Names starting with TEMP or containing TEMPORAL are normalized to "Sin asignar".

Room / Location Codes

UTM room codes follow the pattern CAMPUS-FACULTY-FLOOR-ROOM-TYPE. The parser decodes them into human-readable strings:

1-59-2-04-A

→ Aula 204 - Piso 2 (type code A = classroom)

1-59-3-06-LC

→ Lab. Computación 306 - Piso 3 (type code LC = computer lab)

If the code does not match the UTM pattern, the raw TIPO and LUGAR fields from the schedule block are used instead.

ParseResult Shape

The parsing functions return a ParseResult object consumed by useScheduleUpload to build the final Schedule.

sessions

ClassSession[]

required

Array of parsed class sessions. Each entry includes id, subject, day, startTime, endTime, teacher, location, floor, isVirtual, conflict, and color.

faculty

string

Faculty name extracted from the PDF header (e.g. "FACULTAD DE CIENCIAS INFORMÁTICAS").

academic_period

string

Normalized academic period string (e.g. "SEPTIEMBRE 2025 - ENERO 2026").

student_name

string

Student full name from the ESTUDIANTE: header field.

career

string

Career / school name from the ESCUELA: header field.

Accepted File Types

MIME Type	Processing Path
`application/pdf`	AI extraction → local regex parser fallback
`image/png`, `image/jpeg`, `image/webp`, `image/*`	OCR via Tesseract.js

The DropZone UI currently restricts the file picker to application/pdf only. Image upload is available programmatically via the useScheduleUpload hook for future UI expansion.

Get Started

Core Features

Integrations

Configuration & Deployment

Schedule Upload: Parsing UTM SGU PDFs with Inforario

The DropZone Component

Parsing Pipeline

Metadata Extraction

PERIODO

FACULTAD

ESCUELA

ESTUDIANTE

Normalization Logic

Subject Names

Teacher Names

Room / Location Codes

1-59-2-04-A

1-59-3-06-LC

ParseResult Shape

Accepted File Types

Build docs developers (and LLMs) love

Get Started

Core Features

Integrations

Configuration & Deployment

Documentation Index

​The DropZone Component

​Parsing Pipeline

​Metadata Extraction

PERIODO

FACULTAD

ESCUELA

ESTUDIANTE

​Normalization Logic

​Subject Names

​Teacher Names

​Room / Location Codes

1-59-2-04-A

1-59-3-06-LC

​ParseResult Shape

​Accepted File Types

Build docs developers (and LLMs) love

The DropZone Component

Parsing Pipeline

Metadata Extraction

Normalization Logic

Subject Names

Teacher Names

Room / Location Codes

ParseResult Shape

Accepted File Types