Elliot Bouchy

Situation:

Intern at Archipad, a leader in digital solutions for construction site management, I joined the R&D team to improve the application's user experience through the automation of architectural plan analysis and segmentation.

Task:

My mission: develop and deploy an AI module capable of automatically segmenting PDF plans, extracting textual information, and identifying different rooms, in order to speed up and improve the reliability of document processing for users.

Action:

- Development of advanced segmentation algorithms on vector images, enabling automated plan segmentation with 92% accuracy (measured on an annotated test set).

- Integration of an OCR engine on PDF plans, increasing the correct extraction rate of textual information by 30% compared to the old manual method.

- Implementation of convolutional neural networks (CNNs) for detecting multiple rooms and dwellings, reducing the processing time per plan by 70% (from 10 minutes to 3 minutes on average).

- Application of preprocessing and filtering techniques to improve the quality of extracted data, which reduced the segmentation error rate by 15%.

- Technical watch, rapid prototyping, and validation of technological choices through successive POCs.

- Deployment of the AI module in production via a robust MLOps architecture (Docker, CI/CD, AWS), ensuring scalability, traceability, and real-time monitoring of model performance.

Situation: Quality engineers were losing time on manual tasks during inspections and reporting, hindering overall productivity.

Task: My mission was to digitize and automate these processes through a custom business application.

Actions: Developed an application connected to SharePoint, enabling digitalized quality inspection data entry and tracking. Created VBA scripts to automate the generation and distribution of quality reports, reducing manual interventions. Integrated the application into internal processes, ensuring compatibility with existing APIs and distribution on SharePoint.

Results: Saved 30 minutes per day per quality engineer, amounting to several hundred hours saved over the period. Tangible improvement in the reliability and traceability of field inspections, praised by operational teams.