riasec-co

Connects each student's vocational profile with Colombia's 17,230 active SNIES programs.

Bayesian adaptive RIASEC engine with the complete SNIES catalog — 33 departments, 297 institutions. Zero runtime dependencies.

Install

npm install riasec-co

Quick Start

import { createQuiz, loadPrograms, recommend } from 'riasec-co'

// 1. Run adaptive quiz (~12 questions instead of 48)
const quiz = createQuiz({ language: 'es', mode: 'adaptive' })
while (!quiz.isComplete()) {
  const q = quiz.nextQuestion()!
  const response = await askUser(q.text) // your UI
  quiz.answer(q.id, response)
}

console.log(quiz.profile())
// { R: 0.19, I: 0.27, A: 0.10, S: 0.16, E: 0.10, C: 0.18 }
console.log(quiz.topTypes(2)) // ['I', 'R']

// 2. Get program recommendations
const results = recommend({
  profile: quiz.profile(),
  programs: loadPrograms(),
  filters: { departments: ['Sucre', 'Cordoba', 'Bolivar'], active: true },
  priors: { enrollmentWeight: -0.3, virtualBoost: 1.5 },
  limit: 10,
})

results.forEach((r, i) =>
  console.log(`${i + 1}. ${r.program.nombre_programa} (${r.score.toFixed(2)})`)
)

API

Quiz

| Function | Description | |----------|-------------| | createQuiz(config?) | Create a quiz instance (language: "es" or "en", mode: "adaptive" or "full") | | quiz.nextQuestion() | Next item (or null if complete) | | quiz.answer(id, response) | Record a Likert response (1-5) | | quiz.isComplete() | Whether the quiz has enough information to stop | | quiz.profile() | Current RIASEC profile (RIASECProfile) | | quiz.topTypes(n?) | Top N types sorted by probability | | quiz.progress() | { answered, total, estimatedRemaining, confidence, entropy } |

Data

| Function | Description | |----------|-------------| | loadPrograms() | All 30,809 SNIES programs (17,230 active) | | loadItems(language) | 48 IPIP items in "es" or "en" | | loadMapping() | RIASEC → CINE field weights | | getCINEFields(programs) | Unique CINE broad fields | | getDepartments(programs) | Unique departments |

Recommendations

| Function | Description | |----------|-------------| | recommend(config) | Rank programs by profile fit, with enrollment priors and regional boosts |

Scoring Internals

| Function | Description | |----------|-------------| | uniformPrior() | Dirichlet α = (1,1,1,1,1,1) | | updateAlpha(alpha, type, response, keyed) | Bayesian update after one answer | | posteriorMean(alpha) | E[θ_k] = α_k / Σα | | entropy(alpha) | Shannon entropy in bits | | confidence(alpha) | 1 − (entropy / max_entropy) | | cosineSimilarity(a, b) | Vector similarity [0, 1] | | expectedInfoGain(alpha, type) | Information gain for adaptive item selection |

How It Works

1. Dirichlet prior — starts with uniform belief over 6 RIASEC types
2. Adaptive selection — picks the question that maximizes expected information gain
3. Bayesian update — each Likert response updates the posterior
4. Entropy stopping — stops when Shannon entropy < 1.5 bits (~12 questions)
5. Program matching — cosine similarity between profile and CINE field vectors
6. Enrollment priors — downweights oversaturated fields, boosts underserved ones

Data

Bundled in the package (no network requests):

• 30,809 programs from Colombia's SNIES/HECAA catalog (24 columns each)
• 48 IPIP items in English and Spanish (8 per RIASEC type)
• RIASEC → CINE mapping with weighted associations

Also Available

• Python: pip install riasec-co — PyPI
• R: remotes::install_github("afromero/riasec-co", subdir = "r")
• React components: riasec-co-react

License

MIT