AI Slide Analyzer & Recommender

Summary I built an end-to-end AI system that analyzes PowerPoint slides, detects layout and readability issues, evaluates style consistency, and generates actionable improvement recommendations. The system uses deep-learning computer vision models, a dual-branch CNN for style comparison, and a hybrid recommender engine to suggest tailored improvements.

Problem People often create slides with too much text, bad layout, weak contrast, or inconsistent styles. Manual review is slow, subjective, and hard to scale.

Solution The platform converts each slide to an image, runs multi-label classification to detect design issues, compares styles across the deck with a dual-branch CNN, and provides personalized recommendations. It includes a microservice architecture, custom dataset (900+ annotated slides), and full MLOps monitoring.

Impact Fast, objective slide evaluation without manual review. Users receive clear, actionable feedback and learning suggestions — improving clarity and presentation quality at scale.

Role Sole developer — architecture, models, backend, database, deployment, dataset, and MLOps. This is part of my PHD thesis.

Tech TensorFlow, Python, FastAPI, Docker, PostgreSQL, CNNs, multi-label classification, MLOps.

Vlahova, M., “Recommender Systems: Types, Advantages, Disаdvantantages and Evaluation Techniques,” Journal Computer and Communications Engineering, vol. 15, no. 2, 2021, pp. 3-11, ISSN 1314-2291.

Vlahova, M., “A Scientific Approach to Problem Solving”, Proc. of 1st EUt+ ELaRa Conference “Technologies and Techniques to Support Sustainable Education in the Academic Sphere”, 14-15 December 2022, TU-Sofia

Vlahova, M., Lazarova, M., “Collecting a Custom Database for Image Classification in Recommender Systems,” Proc, of 10th IEEE International Scientific Conference on Computer Science (COMSCI), Sofia, Bulgaria, 30 May 2022 – 2 June 2022, https://doi.org/10.1109/COMSCI55378.2022.9912591 (индексирана в Scopus)

Vlahova–Takova, M., Lazarova, M., “CNN Based Multi-Label Image Classification for Presentation Recommender System,” International Journal on Information Technologies & Security, vol. 16, no. 4, pp. 73-84, 2024, https://doi.org/10.59035/PUYE7368 (индексирана в Scopus и WoS, IF 1.0, Q4)

Vlahova-Takova, M., Lazarova, M., “Dual-Branch Convolutional Neural Network for Image Comparison in Presentation Style Coherence,” Journal Engineering, Technology & Applied Science Research, vol. 15, no. 2, pp. 21719–21727, 2025, https://doi.org/10.48084/etasr.9571 (индексирана в Scopus)

Vlahova-Takova, M., Lazarova, M., “A Recommender System Model for Presentation Advisor Application Based on Multi-Tower Neural Network and Utility-Based Scoring,” Journal MPDI Electronics, vol. 14, 2025, https://doi.org/10.3390/electronics14132528 (индексирана в Scopus и WoS, IF 2.6, Q1)