Hebrew AI course generation proven in 3 months: an EdTech POC ready for MVP
B. Tech needed to know whether generative AI could create and teach full Hebrew courses at a quality worth building on. Rollout IT delivered a working proof of concept in three months, with course generation, quizzes, and an interactive AI tutor, and no reported delays.
Exploring an AI-first learning product? Talk to us
Snapshot
| Field | Detail |
|---|---|
| Client | B. Tech |
| Industry | EdTech / language and course delivery |
| Region | Israel |
| Company stage | Concept-stage product moving toward MVP |
| What we built | AI-first Hebrew course generator and tutor (POC) |
| Tech stack | Next.js, TypeScript, Vercel AI SDK, PostgreSQL, Railway |
| Time-to-value | Research from mid-May 2025, POC delivered August 2025 |
| Headline KPIs | Hebrew feasibility proven – 2 major AI milestones completed – no reported delays |
The challenge
B. Tech came with a product vision, not a platform, and one make-or-break question: if the AI could not generate and teach in Hebrew, no future milestone would matter. Hebrew is underrepresented in many generative AI workflows, so output quality, grammar, cultural nuance, and teaching flow all carried real risk. The POC also needed a functional interface quickly, so the team could test the learning experience itself rather than wait for a polished product shell.
Why B. Tech chose Rollout IT
The client needed a partner comfortable with research-heavy AI work, fast prototyping, and uncertain requirements. Rollout IT treated the first phase as a feasibility and product-learning sprint, using short feedback loops and committing to technical decisions only after testing the AI's actual behavior, which is how you de-risk an unproven idea without burning the budget.
The solution
We began with technical consultation and experiments around Hebrew output quality, prompt structure, and agent orchestration. Once there was enough evidence, we built the POC with a compact two-person core team: a full-stack developer with AI expertise and a technical project manager.
The platform generates course structures, lesson content, quizzes, and exercises in Hebrew. A separate AI instructor agent guides students through the material, answers questions, and adapts to the learner's pace. The agentic flow connects multiple AI calls, using the output of one stage as the input to the next, so content creation and teaching dialogue stay coherent. Next.js and TypeScript supported rapid iteration, while Railway and PostgreSQL kept the POC deployable without heavy infrastructure. Weekly meetings, async updates, and demo videos let the client shape the product in real time.
Results
- The POC proved Hebrew course generation and AI tutoring were feasible.
- Two AI milestones were completed: the course generator and the AI teacher agent.
- The architecture created a practical foundation for MVP development, with no reported delays.
| Metric | Before | After | Timeframe |
|---|---|---|---|
| Product state | Idea, no platform | Working AI-first learning POC | Mid-May to August 2025 |
| Hebrew AI feasibility | Unproven | Course generation and tutoring demonstrated | POC phase |
| Core AI milestones | 0 | 2 completed (generator + tutor) | POC phase |
| Delivery predictability | High research risk | No reported delays | Planned POC window |
The AI angle
AI is both the production engine and the learning interface. The value comes from combining generative course creation with a tutor that teaches dynamically, in a language where off-the-shelf AI performance could not be assumed. Proving that first made every later investment safer.
What's next
The logical next phase is an MVP with learner analytics, stronger course-review workflows, admin controls, and quality evaluation for generated Hebrew lessons across multiple subjects. Rollout IT is positioned to take the proven POC into that build.
FAQ
What did the proof of concept prove? That generative AI can create and teach full Hebrew courses at a usable quality, via a course generator and an interactive AI tutor.
How long did it take? Research started in mid-May 2025 and the POC was delivered in August 2025, roughly three months.
Why was Hebrew a risk? Hebrew is underrepresented in many generative AI workflows, so grammar, nuance, and teaching flow needed to be validated before committing to an MVP.
What was it built with? Next.js, TypeScript, the Vercel AI SDK, PostgreSQL, and Railway.
Validate the hard question first, then build the product around what works. Talk to us.