SGU Schedule Parser — Local PDF and Image Parser Reference

sguRegexParser.ts is Inforario’s fully local, zero-network schedule parser. It processes the official UTM Sistema de Gestión Universitaria (SGU) schedule PDF (or an image scan) entirely inside the browser using pdfjs-dist for PDF text extraction and Tesseract.js for OCR fallback. The output is a structured ParseResult ready to be rendered in the schedule grid. The two public exports are parseScheduleFile — the main entry point — and resolveConflicts — a standalone utility used both here and by the Supabase Edge path.

Types

ParseResult is the local return type for parser output. ClassSession is imported from types/sgu.ts — see the Types Reference page for its full field documentation.

interface ParseResult {
  sessions: ClassSession[];
  faculty?: string;
  academic_period?: string;
  student_name?: string;
  career?: string;
}

// ClassSession is imported from '../../types' (types/sgu.ts)
import { ClassSession } from '../../../types';

Exported functions

`parseScheduleFile()`

parseScheduleFile(base64Data: string, mimeType: string): Promise<ParseResult>

The main entry point for local schedule parsing. Strips the data:*;base64, prefix from base64Data, then routes to the appropriate back-end parser based on mimeType:

application/pdf → parsePDF() (pdfjs-dist positional column parser)
image/* → parseImage() (Tesseract.js OCR → parseRawText())
Anything else → throws "Tipo de archivo no soportado: <mimeType>"

resolveConflicts() is called internally by both paths before the result is returned, so the sessions array in the resolved ParseResult is always conflict-annotated and sorted.

base64Data

string

required

A base64-encoded data URL of the file, e.g. data:application/pdf;base64,JVBERi0x…. The data:*;base64, prefix is stripped automatically before decoding.

mimeType

string

required

MIME type of the uploaded file. Determines which parser is invoked.

Value	Parser used
`application/pdf`	`parsePDF()` via `pdfjs-dist`
`image/png`, `image/jpeg`, …	`parseImage()` via Tesseract.js OCR

Returns Promise<ParseResult>

const result = await parseScheduleFile(base64, 'application/pdf');
// result.sessions    → ClassSession[]
// result.faculty     → "FACULTAD DE CIENCIAS INFORMÁTICAS"
// result.academic_period → "ABRIL 2025 - AGOSTO 2025"
// result.student_name    → "Juan Pérez"
// result.career          → "INGENIERÍA EN SISTEMAS"

Error handling Any error whose message begins with "No se pudo" or "Tipo de archivo" is re-thrown verbatim. All other unexpected errors are wrapped in a generic user-friendly message:

"Error al analizar el horario. Por favor asegúrate de que el archivo sea legible y contenga un horario válido."

`resolveConflicts()`

resolveConflicts(sessions: ClassSession[]): ClassSession[]

A non-destructive conflict detector and sorter. Accepts a flat array of ClassSession objects (possibly mixed schedulable and unscheduled) and returns a new sorted array with conflict flags set correctly. Algorithm

Split sessions into two buckets:
- Schedulable: sessions that have a day, startTime, and endTime
- Unscheduled: virtual classes or sessions missing any time field
Reset conflict = false on all sessions.
Sort schedulable sessions by day (locale string order) then startTime ascending.
O(n²) pairwise scan over same-day sessions. For each pair, convert HH:mm to total minutes and detect overlap with:
```
start1 < end2 && start2 < end1
```
If overlapping, set conflict = true on both sessions.
Return [...schedulable, ...unscheduled] — schedulable sessions first, unscheduled appended at the end.

sessions

ClassSession[]

required

Array of class sessions. May contain any mix of schedulable and unscheduled entries. The original objects are mutated in place (the conflict flag is written directly) but none are removed.

import { resolveConflicts } from './features/uploader/utils/sguRegexParser';

const annotated = resolveConflicts(parsedSessions);
const conflicts = annotated.filter(s => s.conflict);
console.log(`${conflicts.length} conflicting sessions detected`);

resolveConflicts is non-destructive: it never removes sessions. Every session in the input appears exactly once in the output. Only the conflict property is changed.

Constants

`SUBJECT_COLORS`

The ordered palette used to assign deterministic colors to subjects. Colors are assigned round-robin based on the number of distinct subjects already seen, using subject name (uppercased) as the map key.

const SUBJECT_COLORS = [
  '#22C55E',
  '#3B82F6',
  '#F97316',
  '#EF4444',
  '#A855F7',
  '#06B6D4',
  '#EAB308',
];

The helper getSubjectColor(subject, subjectColors) maintains a Map<string, string> across a single parse run. The same subject name always receives the same color within a parse, and the color can be later overridden by the user via the CUSTOMIZE_COLOR feature.

Internal functions

The following functions are not exported. They are documented here as implementation reference.

parsePDF(base64)

Decodes the raw base64 string with atob() (browser) or Buffer.from() (Node.js) and loads the binary into pdfjs-dist. For every page it:

Calls page.getTextContent() and retrieves each item’s text content plus its transform[4] (x) and viewport.height − transform[5] (y) coordinates.
Passes all TextItem objects through extractMetadata() and extractSubjectBlocks().
Calls resolveConflicts() on the resulting sessions.

Column layout constants used by extractSubjectBlocks():

Constant	Value	Purpose
`SUBJECT_MAX_X`	`170`	Right boundary of the subject/asignatura column
`DOCENTE_MIN_X`	`260`	Left boundary of the teacher name column
`DOCENTE_MAX_X`	`415`	Right boundary of the teacher name column
`HORARIO_MIN_X`	`495`	Left boundary of the schedule/location column

Subject names spanning multiple rows are concatenated in y-order. Items within 15 px vertically are grouped into the same subject. The day and time are extracted from strings matching:

/(?:-\s*)?\b(LUNES|MARTES|MI[EÉ]RCOLES|JUEVES|VIERNES)\b\s*\((\d{1,2}):(\d{2}):\d{2}-(\d{1,2}):(\d{2}):\d{2}\)/i

parseImage(base64, mimeType)

Reconstructs a data URL from the raw base64 and MIME type, then calls Tesseract.recognize(imageDataUrl, 'spa') for Spanish OCR. Progress is logged to the console at the recognizing text status.If the recognised text is shorter than 20 characters the function throws:

"No se pudo extraer texto de la imagen. Asegúrate de que sea legible y contenga un horario válido."

The raw OCR string is forwarded to parseRawText(), which applies the same day-time regex on a line-by-line basis. OCR output is inherently noisier than PDF text, so accuracy depends on image resolution and contrast.

extractMetadata(items)

Reads document metadata from the header region of page 1. Only items where page === 1 and y < 210 are considered.For each header item, the text is compared against these label strings:

Label	Stored as	Post-processed with
`PERIODO:`	`academicPeriod`	`normalizeAcademicPeriod()`
`FACULTAD:`	`faculty`	`.toUpperCase()`
`ESCUELA:`	`career`	`.toUpperCase()`
`ESTUDIANTE:`	`studentName`	Raw value
`NIVEL:`	`level`	Raw value

Value discovery: after finding a label item at index i, the function scans forward in the same headerItems array and returns the first item that is on the same row (|Δy| ≤ 5 px) and to the right (x > label.x).

normalizeTeacherName(rawName)

Converts UTM raw teacher name format into a human-readable “Firstname Lastname” string.UTM format: LASTNAME1 LASTNAME2 FIRSTNAME1 [FIRSTNAME2]
Output format: Firstname1 Lastname1Steps:

Returns 'Sin asignar' for empty strings, names beginning with TEMP , or names containing TEMPORAL.
Strips title prefixes matching: ing, lic, dr, dra, msc, mgtr, phd, abg, arq, econ, prof, sr, sra, srta (run twice to catch double titles like “DR. ING.”).
Splits by whitespace into parts:
- ≥ 3 parts → "${capitalize(parts[2])} ${capitalize(parts[0])}" (Firstname Lastname1)
- 2 parts → "${capitalize(parts[1])} ${capitalize(parts[0])}" (reversed)
- 1 part → capitalized as-is

normalizeLocation(codAmb, tipo?)

Parses UTM classroom codes of the form 1-59-FLOOR-ROOM-TYPE into human-readable location strings.Regex: /\d+-\d+-(\d+)-(\d+)-?([\w]*)/

Code segment	Meaning
Group 1	Floor number
Group 2	Room number (zero-padded to 2 digits)
Group 3	Type code (e.g. `LC` for Lab. Computación, `A` for Aula)

Output rules:

Type code LC or tipo containing "laboratorio" → "Lab. Computación {floor}{room} - Piso {floor}"
Default → "Aula {floor}{room} - Piso {floor}"

If the code doesn’t match the UTM pattern (e.g. administrative faculties), the raw code is returned, optionally prefixed with tipo.Examples:

Input	Output
`1-59-2-04-A`	`Aula 204 - Piso 2`
`1-59-3-06-LC`	`Lab. Computación 306 - Piso 3`
`VIRTUAL`	`VIRTUAL` (no match → raw value)

normalizeAcademicPeriod(raw)

Converts verbose SGU period strings into a compact form.Regex:

/([A-ZÁÉÍÓÚÜÑa-záéíóúüñ]+)\s+(?:DE(?:L)?)\s+(\d{4})\s+HASTA\s+([A-ZÁÉÍÓÚÜÑa-záéíóúüñ]+)\s+(?:DE(?:L)?)\s+(\d{4})/i

Input	Output
`ABRIL DE 2026 HASTA AGOSTO DE 2026`	`ABRIL 2026 - AGOSTO 2026`
`SEPTIEMBRE DEL 2025 HASTA ENERO DEL 2026`	`SEPTIEMBRE 2025 - ENERO 2026`

If no match is found, the raw string is returned uppercased.

Edge Functions

Frontend Services

Data Types

SGU Schedule Parser — Local PDF and Image Parser Reference

Types

Exported functions

`parseScheduleFile()`

`resolveConflicts()`

Constants

`SUBJECT_COLORS`

Internal functions

Build docs developers (and LLMs) love

Edge Functions

Frontend Services

Data Types

Documentation Index

​Types

​Exported functions

​parseScheduleFile()

​resolveConflicts()

​Constants

​SUBJECT_COLORS

​Internal functions

Build docs developers (and LLMs) love

Types

Exported functions

`parseScheduleFile()`

`resolveConflicts()`

Constants

`SUBJECT_COLORS`

Internal functions