RetainAI

Decision Intelligence Platform for Employee Retention

RetainAI is an end-to-end Machine Learning, Explainable AI, MLOps, and secure multicloud platform for employee-retention analytics.

Rather than providing only employee attrition prediction, RetainAI combines:

• attrition classification;
• survival and time-to-event analysis;
• SHAP-based global and local explainability;
• reproducible experimentation and artifact tracking;
• API-backed prediction services;
• dashboard-driven decision support;
• validation and monitoring foundations;
• provider-neutral AI-assisted retention guidance.

The product is designed as a Decision Intelligence System for Human Resources: analytical evidence remains reviewable, AI providers remain optional, and high-impact employment decisions remain under human accountability.

Product design, architecture, and AI-assisted development by Hubert Ronald.

Project vision

RetainAI evolves traditional attrition prediction into an explainable, multilingual, and provider-neutral retention-intelligence platform.

The long-term direction is to support workforce planning, retention-strategy design, and responsible HR decision-making without coupling the product to a single cloud or AI provider.

Amazon Bedrock and Google Gemini are treated as interchangeable advisor adapters behind a controlled backend. They are not the product itself, and remain disabled by default until evaluation, quota, privacy, and cost controls are satisfied.

Core capabilities

Capability	Purpose
Attrition classification	Estimate employee attrition risk using supervised models
Survival analysis	Estimate retention duration and time-to-event behavior
Explainable AI	Surface SHAP drivers, feature importance, and evidence
Experiment tracking	Preserve reproducible runs, metrics, and artifacts
Prediction API	Expose controlled inference and explanation contracts
Streamlit dashboard	Present model evidence and decision-support outputs
Multicloud delivery	Run the public dashboard on GCP and the backend on AWS
Retention Advisor foundation	Route validated evidence to optional Gemini or Bedrock adapters
Monitoring roadmap	Add data quality, performance, drift, and validation reports

Architecture overview

RetainAI is a layered Decision Intelligence System that combines reproducible data preparation, attrition classification, survival analysis, SHAP explainability, experiment tracking, controlled API services, and dashboard-driven decision support.

Decision Intelligence core architecture

The core architecture presents the product at a high level, from analytical foundations to decision support and the future advisor layer.

Current foundation and product evolution

The detailed view expands the core architecture and distinguishes capabilities that are already implemented or deployed from those that remain planned, optional, or disabled.

Solid elements represent the current foundation. Dashed elements represent the evolution toward validation reports, monitoring and drift detection, evidence retrieval, provider-neutral AI adapters, and the Retention Advisor.

Architecture details

Current deployed multicloud architecture

The v0.4 runtime is deployed and operational. Google Cloud hosts the public product experience, while AWS owns the controlled backend boundary for authentication, quota, prediction, explainability, and logging.

The browser never receives the backend bearer token.
Cloud Run calls the AWS API from server-side Python only.
Lambda owns authentication, quota, logging, and provider routing.
Gemini and Bedrock remain disabled by default.
Application delivery is separate from Terraform infrastructure delivery.

Runtime architecture details

Live endpoints

Surface	URL
Dashboard	https://retainai.hubertronald.dev
Backend API	https://api.retainai.hubertronald.dev
Portfolio	https://hubertronald.dev

Current release

v0.4.0 establishes the multicloud product foundation:

• branded dashboard and API domains with managed HTTPS;
• Cloud Run dashboard and Lambda container backend;
• Terraform-managed AWS and GCP infrastructure;
• server-side bearer-token authentication;
• DynamoDB request quotas;
• Cloud Run scale-to-zero with a one-instance maximum;
• AI, RAG, and vector providers disabled by default;
• application delivery separated from infrastructure delivery;
• product documentation organized by architectural domain.

Documentation

Area	Guide
Documentation index	docs/README.md
Architecture	docs/architecture/README.md
Dashboard	docs/dashboard/README.md
Data	docs/data/README.md
Exploratory analysis	docs/eda/README.md
Modeling	docs/modeling/README.md
MLOps	docs/mlops/README.md
Multicloud delivery	docs/multicloud/README.md
Prompt engineering	docs/prompts/README.md

Additional project files:

• Changelog
• Contributing
• Citation metadata

Application delivery and data safety

RetainAI separates application promotion, infrastructure management, and data processing into independent operational paths.

Dashboard and backend images are promoted manually from main through GitHub Actions using short-lived cloud identity and immutable image references. Documentation-only changes do not qualify for an application release.

Terraform remains the source of truth for cloud infrastructure, including domains, certificates, IAM, secrets, runtime configuration, quotas, and the future protected S3 data-lake foundation. Routine application releases do not run Terraform or perform S3 data operations.

See:

• Manual multi-cloud application release
• Multi-cloud architecture
• Data architecture
• Architecture decisions

Roadmap

• v0.5 — Monitoring and validation: data quality, model performance, prediction drift, validation reports, and retraining readiness.
• v0.6 — Internationalization and provider abstraction: English and Spanish locale catalogs, structured localized explanations, Gemini adapter, and Bedrock adapter.
• v0.7 — Talent context and evaluation: Resume Intelligence, structured context, explainability reports, evaluation datasets, and bias checks.
• v1.0 — Retention Intelligence: bilingual Retention Advisor, Resume Intelligence, monitoring, drift-aware evaluation, auditable explanations, and secure automated delivery.

Psychometric Intelligence remains outside the committed v1.0 scope until RetainAI has appropriate data, validated instruments, methodological evidence, privacy and legal review, bias analysis, and qualified interpretation.

Product credit

Designed and built with ♥ and AI assistance by Hubert Ronald.

© 2026 RetainAI — All rights reserved.

License

Distributed under the MIT License. See LICENSE for more details.