Building and Deploying Enterprise AutoML Solutions

Researched and productized AutoML algorithms while delivering predictive solutions across 10+ client engagements in Korea, Japan, Vietnam, and Malaysia.

Overview

At AILYS (formerly Solidware), I worked as a Machine Learning Scientist across two connected tracks: algorithm research and product development, and hands-on client-facing solution delivery. I contributed to the development and deployment of DAVinCI LABS, an enterprise AutoML platform used primarily by financial-services organizations.

On the product side, I developed new algorithms and improved existing methods across automated unsupervised learning, deep-learning models, model analysis, automated feature engineering, neural architecture search, and reproducible model-training workflows. On the client side, I worked directly with enterprise teams to translate business objectives into predictive-modeling problems, develop and validate models on their data, interpret results, and support adoption through technical education.

  • Role: Machine Learning Scientist
  • Product: DAVinCI LABS enterprise AutoML platform
  • Industry focus: Financial services and insurance, with additional enterprise use cases
  • Client delivery: 10+ on-site projects across Korea, Japan, Vietnam, and Malaysia
  • Scope: Algorithm research, AutoML product development, predictive modeling, client-side validation, adoption support, and technical education
  • Period: 2017–2020

Product Problem

Enterprise teams wanted to use machine learning for prediction and decision support, but producing a useful model required much more than selecting an algorithm.

Each project involved preparing client data, translating a business KPI into a measurable target, choosing suitable modeling strategies, comparing candidate models, validating results, and explaining how the solution should be used. Rebuilding that workflow through custom code for every engagement made delivery slow and made successful adoption dependent on a small number of specialists.

DAVinCI LABS addressed this gap by bringing data analysis, automated modeling, model comparison, evaluation, and interpretation into a repeatable product workflow. Algorithm and product development therefore had to remain connected to the data constraints, evaluation criteria, and operating requirements observed in client projects.

Algorithm Research and Product Development

I researched, developed, and improved machine-learning algorithms for DAVinCI LABS. The work included automated unsupervised-learning workflows for discovering and analyzing structure in unlabeled data, the development and refinement of deep-learning models, and algorithmic mechanisms for reproducible model results. I also explored neural architecture search as an approach to automating model-design decisions. Related product work addressed recurring AutoML problems in model analysis and automated feature engineering, reflected in three patents to which I contributed.

Research outcomes needed to become reusable product capabilities rather than one-off experiments. I contributed to a shared workflow that allowed enterprise data teams to configure, train, validate, compare, and analyze classical machine-learning and deep-learning models without rebuilding model-specific code for every project.

The research and product-development scope included:

  • Automated unsupervised learning and clustering analysis
  • Deep-learning model development and iterative improvement
  • Algorithmic mechanisms for reproducible model results
  • Neural architecture search
  • Data inspection and problem configuration
  • Automated feature preparation and reusable modeling rules
  • Automated model training and candidate comparison
  • Classical ML and deep-learning model workflows
  • Model evaluation, reporting, and interpretation
  • Reusable project configurations for enterprise delivery

The product workflow kept model choices, configurations, and results inspectable so scientists and client teams could evaluate them against the same objective.

Product Feature Examples

Selected DAVinCI LABS features include data preparation, model configuration, automated training, and predictive analysis.

Automated Modeling. Automatically optimizes candidate algorithms and produces model results in a shared workflow.
Algorithm Selection. Compares available algorithms and composes selected candidates into an ensemble.
Parameter Configuration. Selects training modes, algorithms, and model-specific tuning parameters through one interface.
Data Preparation. Identifies fields that require attention and applies data-filling operations within the product workflow.
Rule Conversion. Transforms data characteristics into reusable rule-based features for modeling.
Quick Model Setup. Creates a baseline model from a selected dataset and optional evaluation data.
Time-Series Prediction. Models patterns that evolve over time for forecasting-oriented problems.
Prediction Simulator. Explores controllable input changes against a defined target outcome.

From Business Problem to Adopted Solution

flowchart TB
    A["Business Objective & KPI"] --> B["Data Validation & Problem Framing"]
    B --> C["Algorithm Research & Model Development"]
    C --> D["AutoML Product Workflow"]
    D --> E["Client-Side Evaluation"]
    E --> F["Deployment · Education · Adoption"]
    F -->|"Observed Needs & Failure Cases"| C

Client work began by defining what a successful prediction meant in the customer’s operating context. I worked with project stakeholders to inspect available data, define targets and validation criteria, train and compare predictive models, and evaluate results against client-defined performance indicators.

The selected model then had to fit the client’s workflow rather than remain an isolated experiment. I supported solution configuration, result interpretation, on-site validation, and technical handoff so client teams could continue using and evaluating the platform after the initial engagement.

Client-side evaluation also exposed data limitations, recurring failure modes, and operational requirements that could not be reproduced from product assumptions alone. Feeding those observations back into algorithm and product development connected research decisions to measurable customer needs.

Establishing Client Adoption Across Markets

I participated in more than ten on-site client projects, primarily across banking, financial services, and insurance. The engagements included organizations such as Shinhan, Hana Bank, AEON, SBI, MetLife Korea, DB Insurance, and TPBank across Korea, Japan, Vietnam, and Malaysia.

Many early-stage engagements were structured as hands-on pilot projects so prospective client teams could evaluate DAVinCI LABS through actual use. I worked on site for periods ranging from about one week to two months, collaborating with client teams on a concrete business problem with the goal of establishing a successful first use case on their own data.

This required more than demonstrating a product workflow. The platform had to be validated against each client’s data conventions, infrastructure, modeling expectations, and decision processes. I worked alongside local teams through problem framing, modeling, validation, troubleshooting, and technical enablement.

My responsibilities across these projects included:

  • Running time-bounded on-site pilot projects around concrete business objectives
  • Translating business objectives into measurable prediction tasks
  • Building and training models on client data
  • Comparing models against client-defined evaluation criteria
  • Investigating data and modeling failures during on-site validation
  • Adapting the solution-delivery process to different organizations and markets
  • Delivering technical education so client teams could use and assess the platform

Connecting Research, Product, and Client Delivery

Recurring problems identified through product development and client delivery informed research priorities across reusable AutoML methods.

Three patents reflect my contributions in this area:

  • 10-2273868: Unsupervised-learning technology
  • 10-2273867: Model-analysis technology
  • 10-1976689: Automated feature-engineering technology

Product and Client Impact

Predictive-model delivery through DAVinCI LABS spanned more than ten client engagements and supported the platform’s initial adoption in multiple international organizations.

Client teams received models evaluated against their own objectives as well as the technical context needed to continue using and assessing the platform.

Technical Summary

  • Algorithm and model research: Automated unsupervised learning and clustering, deep-learning model development and improvement, and neural architecture search
  • AutoML productization: Automated feature engineering, model training and candidate comparison, evaluation and reporting, and inspectable project configurations for repeatable workflows
  • Applied predictive modeling: Business and KPI framing, data validation and preparation, target definition, model training, client-defined evaluation, and failure analysis
  • Enterprise pilots and adoption: 10+ on-site engagements across Korea, Japan, Vietnam, and Malaysia, including pilots lasting from approximately one week to two months, with troubleshooting, result interpretation, technical education, and handoff
  • Research outputs: Three patents covering unsupervised learning, model analysis, and automated feature engineering

Public Product

Visit DAVinCI LABS