AI-Enhanced Compliance LMS Platform for Construction & Safety Training

Challenge

The client had an existing LMS model that allowed users to purchase and complete compliance training courses online. However, the system delivered static learning paths and did not adapt to individual learner performance.

From a business perspective, this created several issues:

learners who failed quizzes received generic remediation;
there was no structured way to identify knowledge gaps;
learning plans were not personalized;
manual instructor intervention was required in some cases;
the system lacked intelligent content retrieval across large training materials;
regulatory compliance required full tracking and auditability.

The client needed to build a scalable cloud-native LMS that could:

track quiz performance in detail,
identify weak knowledge areas,
generate adaptive learning plans,
use AI safely within a compliance-driven environment, and
maintain a secure and auditable infrastructure.

Solution We Delivered

We enhanced the client’s existing LMS by integrating an AI-powered personalization layer using Retrieval-Augmented Generation (RAG).

The core LMS platform (an Angular frontend and a Spring Boot backend deployed on AWS) was already operational. Our role was to extend its capabilities by introducing intelligent remediation, knowledge retrieval, and adaptive learning features, without disrupting production workflows.

AI & RAG Integration Architecture

We designed and implemented an AI augmentation layer that integrates seamlessly with the existing backend services.

The architecture includes:

AWS Kendra is indexing all structured and unstructured learning materials.
Amazon Bedrock for LLM inference with Claude 3.7 Sonnet (LLM model used for learning plan generation).
Extended Spring Boot services to orchestrate retrieval and generation using langchain4j.
PostgreSQL schema extensions to store learner performance data and AI-generated plans.

No LMS replacement or core re-platforming was performed. The AI services were integrated as modular components within the existing system.

Adaptive Learning Workflow

We introduced a post-quiz intelligence workflow that operates as an extension of the LMS assessment engine:

A learner completes a quiz within the LMS.
The system evaluates answers at the question level.
Incorrect responses are mapped to defined knowledge domains.
The backend queries AWS Kendra to retrieve relevant learning materials.
The retrieved context is sent to an LLM hosted on Amazon Bedrock.
The LLM generates a personalized remediation plan, including: targeted content recommendations, concept explanations, reinforcement exercises, and suggested learning order.
The generated learning plan is saved in PostgreSQL and linked to the learner profile.
Future quiz attempts update and refine the learning plan dynamically.

This creates a closed feedback loop where learner performance continuously improves personalization accuracy.

LLM Testing and Validation (Prompt Evaluation Framework)

Because the platform operates in a regulated training environment, validating AI outputs was critical.

We implemented a structured LLM evaluation process using Promptfoo to systematically test:

prompt behaviour across multiple scenarios,
output consistency and grounding accuracy,
hallucination risk,
alignment with indexed training materials, and
tone and compliance suitability.

This ensured that the generated remediation plans remained reliable, grounded in Kendra-retrieved content, and suitable for a compliance-driven environment.

Client’s Results

The enhanced LMS platform evolved from a static course delivery system into an adaptive, AI-augmented compliance training environment — without replacing the existing core infrastructure. The platform now delivers:

Personalized Remediation at Scale

The client can now deliver personalized learning plans at scale. Learners benefit from quiz-level, targeted feedback that identifies knowledge gaps, while instructors see improved efficiency and better client training program results.

Closed-Loop Learning Optimization

Each quiz attempt refines the learner’s remediation plan. This continuous feedback loop improves personalization accuracy over time and creates measurable visibility into learner progress at the domain level.

Reduced Instructor Intervention

With AI-driven remediation, the client reduces instructor workload for routine cases. This enables staff to dedicate their expertise to more complex or high-impact learner support, improving operational efficiency.

Enterprise-Grade AI Governance

Through structured LLM validation and grounding via AWS Kendra, AI-generated outputs remain aligned with approved training materials. The platform operates safely within a regulated, audit-driven compliance environment.

Secure, Cloud-Native Scalability

The modular RAG-based architecture lets the client scale securely on AWS, with full auditing and access controls, supporting additional growth in compliance training without operational risk.