Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/diazdavilajesus16-stack/IA-LUMINA/llms.txt

Use this file to discover all available pages before exploring further.

Lumina AI’s knowledge is defined entirely by the intents in respuestas.json. Each intent describes a topic the chatbot recognizes — the example phrases students might use to raise that topic, and the replies Lumina AI can give. Editing this file is the primary way to expand, restrict, or refine what the chatbot understands and how it communicates.

​
File format

respuestas.json contains a single top-level key, "intents", whose value is an array of intent objects. Every object in the array follows the same three-field structure: 
{
  "intents": [
    { "tag": "...", "patterns": [...], "responses": [...] },
    ...
  ]
}

​
Example intent

The following is the algebra_ecuaciones intent taken directly from respuestas.json: 
{
  "tag": "algebra_ecuaciones",
  "patterns": [
    "Ecuaciones",
    "Despejar x",
    "Resolver ecuación",
    "Ecuación de primer grado",
    "Ecuación lineal",
    "Cómo despejar"
  ],
  "responses": [
    "Resolver una ecuación de primer grado es como despejar una incógnita. Ejemplo: 2x + 5 = 13. Paso 1: restas 5 a ambos lados → 2x = 8. Paso 2: divides entre 2 → x = 4. La clave es hacer la misma operación en ambos lados de la igualdad. ¿Tienes un ejercicio concreto?",
    "Imagina que la ecuación es una balanza equilibrada. Si sumas, restas, multiplicas o divides lo mismo en ambos platillos, la balanza sigue equilibrada. Eso es despejar. Dame un problema y lo resolvemos juntos paso a paso."
  ]
}

​
Field descriptions

​
tag
string
required
A unique string identifier for this intent. Must not duplicate any other tag in the array. The tag is used internally to match a prediction to the correct set of responses and to power the keyword fallback map in chatbot.py.
​
patterns
string[]
required
An array of example phrases that a student might type to trigger this intent. These are tokenized and lemmatized during training to build the bag-of-words vocabulary. More varied patterns — different sentence lengths, synonyms, and phrasings — lead to better recognition accuracy.
​
responses
string[]
required
An array of possible replies. When the intent is matched, Lumina AI selects one response at random using random.choice(). Providing multiple responses gives the chatbot variety so it does not feel repetitive across a conversation.

​
How intent matching works

When a student sends a message, Lumina AI processes it through a three-stage cascade defined in chatbot.py:
  1. Neural network prediction — the message is tokenized and lemmatized, then converted to a binary bag-of-words vector over the full training vocabulary. The Keras model (chatbot_model.h5) outputs a confidence score for every intent class. If any class exceeds _THRESHOLD_HIGH = 0.60, that intent is used directly.
  2. Keyword fallback — if no class clears the high threshold, chatbot.py checks _KEYWORD_MAP, a hardcoded dictionary of keywords per tag. If one or more keywords are found in the normalized message, the highest-scoring tag is returned with a synthetic confidence of 0.50.
  3. Low-confidence fallback — if neither stage produces a result, intents that exceed _THRESHOLD_LOW = 0.35 are considered. If still nothing matches, a contextual fallback message is returned.
Because the neural network is trained on the patterns arrays, the more varied and representative your patterns are, the better the model’s accuracy. Aim for at least five to eight distinct phrasings per intent, covering different vocabulary and sentence structures a real student might use.

​
Adding a new intent

1

Open respuestas.json

Open respuestas.json in any text editor. Locate the closing ] of the "intents" array (the last intent ends with } before it).
2

Add a new intent object

Insert a comma after the last intent’s closing }, then add your new object. Choose a unique tag, write at least five to eight varied patterns, and provide two or more responses so the bot has variety:
{
  "tag": "estadistica_basica",
  "patterns": [
    "Qué es la estadística",
    "Media aritmética",
    "Cómo calcular la mediana",
    "Moda de un conjunto de datos",
    "Medidas de tendencia central",
    "Varianza y desviación estándar",
    "Estadística descriptiva",
    "Datos y frecuencias"
  ],
  "responses": [
    "La estadística descriptiva resume datos con medidas de tendencia central: la media (promedio), la mediana (valor central al ordenar los datos) y la moda (valor más frecuente). La varianza mide la dispersión; la desviación estándar es su raíz cuadrada. Dame un conjunto de números y calculamos todo juntos.",
    "Para calcular la media, suma todos los valores y divide entre la cantidad. La mediana requiere ordenar los datos primero. La moda es el valor que más veces aparece. ¿Qué medida necesitas calcular?"
  ]
}
3

Save the file

Save respuestas.json. Make sure the JSON is valid — each intent object is separated by a comma, and the overall structure is { "intents": [ ... ] }.
4

Retrain the model

Run the training script from the project root. This re-reads all patterns, rebuilds the vocabulary, and overwrites chatbot_model.h5, words.pkl, and classes.pkl:
python training_chatbot.py
5

Restart the app

Start Lumina AI again so it loads the updated model and intent file:
streamlit run Main.py

​
Existing intent tags

The following tags are defined in chatbot.py’s _KEYWORD_MAP and are used as the keyword-fallback targets. Each corresponds to one or more intents in respuestas.json:
TagTopic
saludoGreetings
despedidaFarewells
matematicas_generalGeneral mathematics
algebraAlgebra (equations, polynomials, factoring)
fraccionesFractions
geometriaGeometry (shapes, area, volume)
trigonometriaTrigonometry
probabilidadProbability and statistics
biologiaBiology
quimicaChemistry
fisicaPhysics
historiaHistory
literaturaLiterature
programacionProgramming
geografiaGeography
tecnicas_estudioStudy techniques
motivacionMotivation
habitosStudy habits
examenExams and tests
concentracionFocus and concentration
curiosidadesFun facts and curiosities
agradecimientoGratitude and thanks
chisteJokes and humor
ansiedad_academicaAcademic anxiety
salud_mental_estudioMental health while studying
no_entendidoClarification requests
Adding or editing intents requires retraining the model. Running python training_chatbot.py overwrites chatbot_model.h5, words.pkl, and classes.pkl with files trained on the current contents of respuestas.json. The previous model is not backed up automatically — if you want to keep it, copy chatbot_model.h5 before retraining.

Build docs developers (and LLMs) love

Get started for freeTalk to us