DELE AI Exam Platform
An intelligent SaaS platform for creating, delivering, and evaluating Spanish DELE exam practice with CEFR-aware AI workflows.
About the product
The DELE AI Exam Platform is a SaaS product designed to generate, publish, complete, and evaluate practice examinations for Spanish-language DELE certification.
Our role
We designed the product, its exam-domain data model, the full-stack architecture, and the AI generation and evaluation workflows. The work spans product research, backend services, user-facing examination flows, quality-control systems, and SaaS infrastructure.
How we approached it
Official DELE structures were translated into reusable levels, modalities, sections, task types, scoring rules, and templates. AI generation was then placed inside a controlled pipeline: prompts receive CEFR and task constraints, outputs pass through language and structural validators, invalid results can be repaired, and generated material can be compared with patterns extracted from official-style corpora.
AI-assisted exam generation
The platform uses large language models to generate exam texts, tasks, questions, answer options, and distractors. Generation is controlled by exam level, skill, task structure, word limits, and CEFR requirements rather than relying on unrestricted prompts.
Official-exam alignment
An analysis pipeline studies official-style examination material and builds reusable corpus patterns. Generated content can then be compared with those patterns to improve structure, vocabulary, question style, difficulty, and source-text alignment.
Validation and quality control
The AI workflow includes validation layers for Spanish output, malformed options, repeated answers, missing content, inappropriate language leakage, and alignment between questions and their source material. Invalid outputs can be repaired or sent through deterministic fallback logic.
Evaluation system
The product architecture supports user submissions, section and task results, scoring groups, grading thresholds, and AI-assisted evaluation workflows. Writing components include level-aware analysis and calibration foundations for automated feedback.
SaaS and content architecture
Beyond AI generation, the system includes reusable exam templates, levels, modalities, sections, task types, content blocks, media assets, published exam versions, authentication, user progress, subscriptions, and billing infrastructure.
Technology context
The platform combines Next.js, TypeScript, AdonisJS, relational data models, Redis, OpenAI and Anthropic integrations, NLP services, Stripe, and S3-compatible media infrastructure.
This is one of our own products or internally developed projects.