B. Tech is an edtech company based in Israel that aims to reinvent course development and delivery.

Its dream is to craft an AI-first learning space that can generate full courses in Hebrew, a largely under-represented language by existing generative AI solutions and disseminate them through an interactive AI tutor.

This is not just content creation. “The AI-powered LMS builds complete course models with lessons and quizzes on any subject from cooking to career education and then teaches the content in real time to the students.”The AI “educator” adjusts dynamically to the student’s pace and questions, offering a highly individualized learning experience.

The Challenge

When B. Tech approached us, they had no platform, but an idea. Proof of Concept (PoC) had one make-or-break condition:

If the AI couldn’t generate and teach in Hebrew, there wouldn’t be future project milestones implemented.

The key challenges were:

• Proving high-quality generative AI output in Hebrew, bridging language and cultural gaps in existing models.
• Creating a sophisticated agentic flow where multiple AI calls notify each other, keeping context and instructional cohesion.
• Implementing a functional UI in the shortest possible time for proof of concept testing, sacrificing visual quality for speed and feasibility.
• Operating in a research-intensive space in which many technical decisions had to be discovered by trial and error rather than established specs.

Our Solution

We started with deep technical consultation and research, doing a series of experiments on Hebrew output quality and agent orchestration approaches prior to making architectural commitments.

We utilized an iterative POC build process where the customer was able to see changes in real time, enabling instant feedback loops and feature pivots.

AI-Centric Features Delivered:

• AI course generator: Creates fully formatted courses in Hebrew, including lesson outlines, complete content and exercises.
• AI Instructor Agent: Guides students through the generated course step-by-step, answering questions, explaining terms and altering the lesson flow.
• Advanced Agentic Flow: Manages multi-stage AI inference where an AI result from one stage becomes input to the subsequent stage — from content generation to teaching dialogue.
• Hebrew Language Optimisation: Advanced prompt engineering and tuning methods for handling Hebrew syntax, grammar and idiomatic expression.

Tech decisions

• Frontend & Backend: Next.js for common development experience and rapid iteration.
• Infrastructure: Railway hosting with PostgreSQL database for rapid POC deployment.
• Generative AI: Using custom Vercel AI SDK to enable flexible orchestration of multi-step AI calls.
• TypeScript for type safety and code maintainability.

Collaboration & Team Structure

Project developed by a two-member core team:

• Full-Stack Developer & AI Expert
• Technical Project Manager

We had weekly meetings complemented with ongoing asynchronous communication and demo videos. The client was very involved in decision-making, co-designing the product in real-time.Timeline & Milestones

• Mid-May 2025: Start for research and AI prototyping.
• Milestone 1: Functional Hebrew AI Course Generator.
• Milestone 2: Complete AI Teacher Agent with built-in agentic flow.
• August 2025: POC delivery.

No delays were incurred – the AI and Hebrew problems were solved within the planned timeframe.

Results

Met all POC goals: Hebrew AI course development and teaching were entirely possible.

Schemed a scalable architecture for future MVP and beyond.

Received extremely positive client response – the AI quality and interactivity were above expectations.

Lessons Learned

Underrepresented languages’ generative AI demands more amounts of research and iterative fine-tuning than English-based solutions. Educational agent orchestration requires accuracy of content, flow and conversational sense pedagogically — and hence AI prompt engineering and flow designing are critical engineering tasks.

The AI LMS project shows how generative AI and agentic architectures can redefine digital learning, even in linguistically challenging markets like Hebrew. By proving that AI can generate and teach with the nuance of a human instructor, we’ve laid the foundation for a platform that could scale across subjects, audiences and geographies — starting with a groundbreaking POC.