§ Research

Applying what we research.

aGo lab combines engineering practice with formal academic research. The lab's director is pursuing a joint doctorate (University of Málaga, Spain and University of Bío Bío, Chile) on predictive analysis of cultural consumption. The techniques validated in our publications are applied as a methodological foundation to the predictive problems we address for clients.

§ 01 · peer-reviewed publication

Paper in Academia Revista Latinoamericana de Administración.

Open Access Q3 Scopus Emerald May 2025

Unveiling economic barriers to cultural consumption in Chile: insights from SHAP analysis and predictive modeling

Sixto Valdés-Elizalde, Rodolfo Mendoza-Llanos, Sebastian Molinillo, Pedro G. Campos

Academia Revista Latinoamericana de Administración · 38(1) · 53 · 2025

Abstract

Study on how economic, subjective and cultural-capital factors impact cultural-consumption decisions in Chile, using data from the National Cultural Participation Survey. We apply LightGBM as predictive model and SHAP for interpretability. Joint research with the University of Bío Bío and the University of Málaga.

Methods applied

LightGBM: gradient boosting for prediction of consumption patterns.
SHAP (SHapley Additive exPlanations): model interpretability to identify the real drivers of each decision.
Chile ENPC: population dataset N=12,151, processed and validated.

Read full paper on Emerald DOI: 10.1108/ARLA-07-2024-0147

§ 02 · what we are working on today

Active technologies and problems.

Areas we are working on right now. Some have research under peer review and are described in general terms to avoid compromising the editorial process. Others are in production with clients. Others live in stealth.

Research + MVP Under editorial review

Algorithmic accessibility in optimization systems

We study how AI-based recommendation and planning systems do (or do not) model users' functional capabilities: reduced mobility, sensory profiles, accumulated fatigue on physical routes. A research paper is currently under peer review at an indexed international journal, co-authored with a software engineering research team at a European university, and an MVP is being built that applies the findings to an urban pilot case.

relevant stack · Django · PostgreSQL/PostGIS · OpenRouteService · VROOM · Next.js · WCAG AAA

↳ status: paper under editorial review, MVP in construction

In production Regulated verticals

RAG and AI agents for vertical compliance

Assistants with retrieval augmented generation over technical and legal regulations, source citation, intelligent task assignment (for example, auditors) and AI agents for routine administrative tasks. Applied in production to organic certification (Bioaudita client) and to SME accounting (ContabilidadVV client).

relevant stack · Django · embeddings · Gemini · Chilean Law 19.799 · pyHanko

↳ status: in production, iterating with user feedback

In production Open data

Open statistical analysis with differential privacy

Analytical portals over public data with formal privacy protocols: k-anonymity (k≥5), concentration indices (HHI, Gini), association tests with familywise error correction (Holm-Bonferroni), Lorenz curves, and cryptographic provenance traceability. Applied to the Chilean organic sector on SAG public data. CC-BY 4.0 license.

relevant stack · Django · PostgreSQL · bootstrap statistical inference · SHA256

↳ status: in production, datalab.bioaudita.cl

Stealth In-house R&D

Predictive modeling and contextual recommendation

Proprietary initiatives around dynamic-context-aware cultural recommendation and interpretable predictive models on vertical domains. Methodological foundation: LightGBM + SHAP, validated in the paper published in Emerald. Product and architecture details are not made public.

relevant stack · LightGBM · SHAP · embeddings · hybrid architectures

↳ status: stealth, conversations under NDA

§ 03 · applied research lines

From academia to product.

These are the techniques we use when a client brings us a predictive or explanatory problem.

 Predictive models with interpretabilityLightGBM, XGBoost, penalized logistic regression. Interpretability with SHAP so the model is never a black box. Used in projects measuring cultural impact and analyzing consumption.
 Retrieval Augmented Generation (RAG)Queries over technical and legal regulations with source citation. Deployed in production for organic certification (Bioaudita) and accounting compliance.
 Conversational AI assistantsIntegration with provider models (Gemini, OpenRouter, Cerebras) for documentary assistance, automatic file validation, and AI agents for routine administrative tasks.
 Privacy and statistical rigorK-anonymity (k≥5), χ² tests with Holm-Bonferroni correction, HHI/Gini/Lorenz indices, robust deduplication and SHA256 traceability. Applied to Bioaudita's DataLab over public SAG data.
 

§ 04 · doctoral research in progress

Doctoral thesis.

Predictive analysis of cultural goods consumption using machine learning

Doctorate in Economics and Business · University of Málaga · Joint program with University of Bío Bío · 2023 to present

Research on the factors that affect cultural consumption using machine learning, aimed at developing predictive models that improve cultural planning. A methodological line that connects with the published paper and with the impact-measurement system that runs at Data Cultura.

§ FAQ

About our research

If your question isn't here, email us at hola@ago.cl or via WhatsApp.

01 Who leads aGo lab's research line?

Sixto Valdés Elizalde, founder and technical director of aGo lab. Joint doctoral candidate between the University of Málaga, Spain (Economics and Business) and the University of Bío Bío, Chile (Economics and Information Management). He is also pursuing a Master in Software Development at the University of Granada.

02 What scientific publications does aGo lab have?

Paper published in Academia Revista Latinoamericana de Administración (Emerald, 2025): Valdés-Elizalde, S., Mendoza-Llanos, R., Molinillo, S., and Campos, P. G. Unveiling economic barriers to cultural consumption in Chile: insights from SHAP analysis and predictive modeling. 38(1), 53. Open Access. DOI: 10.1108/ARLA-07-2024-0147.

03 What machine learning techniques does aGo lab use?

LightGBM and other gradient boosting methods for predictive modeling. SHAP for interpretability. RAG (Retrieval Augmented Generation) over technical and legal regulations with source citation. Conversational AI agents on Google Gemini, OpenRouter, Cerebras. Inferential statistics with k-anonymity, HHI/Gini/Lorenz indices and χ² tests with Holm-Bonferroni correction.

04 What problems is aGo lab working on right now?

Four active areas: algorithmic accessibility in optimization systems (research under peer review), RAG and AI agents for vertical compliance in organic certification and accounting, open statistical analysis with differential privacy on public data, and predictive modeling applied to contextual cultural recommendation.

05 Where can I read the published paper?

On Emerald, Open Access: https://www.emerald.com/arla/article/38/1/53/1274532/Unveiling-economic-barriers-to-cultural

§ next

Does your problem need applied research?

Prediction, interpretability, RAG, rigorous statistical analysis. If what you need looks like this, let's talk.

Start a conversation