AICET — Artificial Intelligence Competency Evaluation Test

🎯 The Objective

The goal of AICET is to design a trusted, standardized AI test assessing the competencies of engineers, researchers, and professionals working with artificial intelligence. The test must cover the full spectrum of AI knowledge — from ethics and data governance to algorithmic principles, coding, and model evaluation — while maintaining high scientific integrity.

To achieve this, the consortium developed an automated question generation and validation pipeline capable of producing and filtering thousands of questions, all linked to a well-defined ontology and validated by domain experts before certification.

💡 The AICET Pipeline Concept

BionomeeX was responsible for creating the data generation and management pipeline, the heart of the AICET system. The pipeline automates the generation of AI-related questions using Large Language Models (LLMs) and ensures their consistency through a multi-step validation workflow.

1. 🧠 Question Generation: AI models generate candidate questions following a structured ontology.
2. 🏷️ Ontology Tagging: Each question is annotated with multiple tags (topic, difficulty, skill type).
3. 🔍 Automatic Filtering: Scripts parse and clean malformed, redundant, or ambiguous questions.
4. 👩‍🏫 Expert Review: Domain specialists validate or reject each question for accuracy and clarity.
5. 📦 Dataset Consolidation: Approved questions are sent to Numalis for integration and test design.

This approach guarantees diversity, balance, and pedagogical validity in the dataset while dramatically accelerating content creation compared to manual methods.

⚙️ Technical Implementation

• Backend: Python + FastAPI API to manage question pipelines and ontology tags
• LLM Integration: Automated generation through OpenAI-compatible endpoints with structured prompts
• Parsing & Filtering: Custom text parsers to detect format errors, duplicates, and incomplete questions
• Validation Tools: Internal review interface for experts to rate and comment on question quality
• Deployment: Dockerized microservices architecture allowing concurrent workflows

The architecture was designed for scalability and transparency, allowing real-time monitoring of data flow from generation to validation and export to Numalis.

📊 Impact & Results

The pipeline successfully produced a large and diverse dataset of AI-related questions, spanning technical, ethical, and applied domains. The system is capable of generating and processing over 1,000 questions per week per engineer at full time, with automated quality checks ensuring coherence and coverage across ontology categories.

• ⚡ 1,000+ questions generated and curated
• 🏷️ 5 ontology trees for fine-grained classification
• 👩‍🏫 Multi-expert validation workflow integrated
• 📈 Scalable, reproducible generation pipeline

This dataset forms the foundation of the official AICET certification managed by Numalis, which will be used nationwide by professional institutions.

🧠 My Role

As part of the BionomeeX engineering team, I was responsible for developing the backend and automation pipeline that connects LLM generation, ontology tagging, and expert validation. My focus was on ensuring scalability, consistency, and maintainability in a rapidly evolving, multi-partner environment.

• 🧩 Designed and implemented the multi-stage generation pipeline in Python
• ⚙️ Developed parsing and validation tools to clean and standardize question outputs
• 🔗 Created APIs for communication between LLM generation services and expert dashboards
• 🧠 Worked on ontology-based tagging and automated quality metrics
• 🤝 Coordinated integration with Numalis and academic reviewers

This project strengthened my skills in data engineering, prompt design, API development, and collaborative software design across institutional and industrial partners.