AI-Native Corporate Learning Platform

The challenge

The client sells corporate training, and two things were holding the business back.

• Creating content was slow and expensive. Experts spent weeks writing each course, and producing videos cost even more.
• Proving that training worked was manual. Staff graded written answers by hand and checked certificates against attendance sheets.

Standard learning platforms couldn't create content, and an earlier in-house attempt had failed because it couldn't keep each client's data separate or control AI costs. My job was to build it properly: a platform where AI does the production work, while still behaving like a dependable, compliant enterprise system.

What I delivered

• A complete learning platform: courses, lessons, classes and schedules, attendance tracking, quizzes and exams, automatic certificates, and scheduled reports.
• AI course writing: a manager describes a training need in one sentence, and the platform plans the course, writes every lesson and creates a quiz that matches it. It draws on each client's own documents and terminology, so the content sounds like them.
• AI video production: each lesson can become a narrated, illustrated training video with no film crew. The AI writes the script, creates the visuals, adds the voice-over and checks the finished video before anyone sees it.
• AI grading and coaching: written answers are graded against a clear rubric, with the manager's decision always final. Learners also get an AI tutor based on each lesson, and managers see skill gaps and learners who may be falling behind.
• An AI interviewer: a voice-based practice interviewer with live subtitles and a scored report.

How it works

Many client organisations use the same platform, so keeping their data apart was the first priority. I enforced that separation in the database itself, not just in the application, so a single coding mistake can't expose one client's data to another.

AI content moves through a production line. Plan, research, draft, check and improve each happen as separate steps, and there is a simpler fast path as a backup. Video production works the same way, and it saves its progress between steps, so nothing is lost if something fails halfway.

Key decisions

• Cost control built into the structure, not added later. Every AI request is measured, and spending is capped per task, per client each month and per user. A double-click can never charge twice. Choosing the right model for each task and cutting wasteful retries brought a complete course to about one dollar.
• Arabic from day one. Right-to-left layouts, correct Arabic lettering on certificates, and English and Arabic always kept in step.
• Keep it simple to run. The client had no dedicated operations team, so I built one well-organised core system with only a few separate services where they were truly needed. I even moved one service back into the core when it added complexity without a real benefit.

Built for trust

• Each client's data is kept separate at the database level, and every new release is automatically checked to make sure that protection is still in place.
• Personal information is removed before content goes into the video pipeline.
• The database, cache, file storage and secrets are never exposed to the public internet.
• AI-generated content goes through an automatic quality check covering structure, accuracy of sources, correct terminology and readability before it's released.
• AI-made videos that fail the visual check wait for a person to approve them.
• The platform can be deployed in the client's preferred cloud region to meet data residency needs.

The results

The client can now create a full course from a single sentence, produce training videos without a production team at a cost they control, grade written answers at scale while managers keep the final say, see exactly where AI budget is going, and run everything in two languages.