Create a 15-slide engineering internship defense deck. Dark tech theme (blue/cyan), minimal text, max 4 short bullets per slide, icons and diagrams. English. Title: Agentic AI for Intelligent Supervision of an Industrial System. Mohamed Louai Darguech, ENSI. ABS Computer, Sfax. Supervisor: Alaeddine Laadhar. 15/06–31/07/2025. 1. Title 2. Introduction: ABS Computer (software publisher since 1993, textile solutions TextiTime/Equitex). Production monitoring lacks an AI layer using real-time data. 3. Context & objectives: Problems: no natural-language data access, no automatic anomaly detection, no predictions. Goal: agentic AI layer complementing existing tools. Constraints: read-only DB, role-based HR access, traceability, no ML leakage. 4. Related works: Industry 4.0, predictive maintenance, LLM agents + RAG. Our approach: IoT + ML + RAG agent in one modular platform. 5. Architecture (diagram): RFID/ESP32 → MQTT → MySQL → Database Engine (PostgreSQL) → ML Engine + RAG + LLM Engine (FastAPI, Gemini) → React frontend. 6. Database Engine: 40 Realtime tables, 68 Parquet datasets, 9 governance tables (117 objects); checkpoints; read-only source. 7. ML Engine: 20 models (duration, anomalies, delays, bottlenecks, maintenance); scikit-learn, XGBoost; temporal splits, leakage fixed. 8. LLM Engine & RAG: FastAPI, Gemini, tool router; 6 read-only SQL views; HR guard (HTTP 403); traceability panel. 9. Frontend: React, TypeScript, Vite; AI Lab chat, forms dashboard. 10. Results – Tests: Frontend 28/28; LLM Engine 117 passed, 2 failed; DB validator 112 OK, 0 errors; simulated P95 24.58 ms. 11. Results – ML: ML01 R²=0.873; ML09 R²=0.8165; ML16 rejected; ML19/ML20 proxy (no real data). Table: validated / proxy / rejected. 12. Discussion: Strengths: modularity, security, honest evaluation. Limits: no live validation with real PostgreSQL/Gemini/ABS data, 14 data contracts missing, no push alerts. 13. Demo: screenshots placeholder (landing, AI Lab, forms). 14. Conclusion: functional, statically validated. Next: real data, live Gemini, notifications. Skills: PostgreSQL, MQTT, FastAPI, React, critical ML evaluation. 15. Questions
Create a 15-slide engineering internship defense deck. Dark tech theme (blue/cyan), minimal text, max 4 short bullets per slide, icons and diagrams. English.
Title: Agentic AI for Intelligent Supervision of an Industrial System. Mohamed Louai Darguech, ENSI. ABS Computer, Sfax. Supervisor: Alaeddine Laadhar. 15/06–31/07/2025.
1. Title
2. Introduction: ABS Computer (software publisher since 1993, textile solutions TextiTime/Equitex). Production monitoring lacks an AI layer using real-time data.
3. Context & objectives: Problems: no natural-language data access, no automatic anomaly detection, no predictions. Goal: agentic AI layer complementing existing tools. Constraints: read-only DB, role-based HR access, traceability, no ML leakage.
4. Related works: Industry 4.0, predictive maintenance, LLM agents + RAG. Our approach: IoT + ML + RAG agent in one modular platform.
5. Architecture (diagram): RFID/ESP32 → MQTT → MySQL → Database Engine (PostgreSQL) → ML Engine + RAG + LLM Engine (FastAPI, Gemini) → React frontend.
6. Database Engine: 40 Realtime tables, 68 Parquet datasets, 9 governance tables (117 objects); checkpoints; read-only source.
7. ML Engine: 20 models (duration, anomalies, delays, bottlenecks, maintenance); scikit-learn, XGBoost; temporal splits, leakage fixed.
8. LLM Engine & RAG: FastAPI, Gemini, tool router; 6 read-only SQL views; HR guard (HTTP 403); traceability panel.
9. Frontend: React, TypeScript, Vite; AI Lab chat, forms dashboard.
10. Results – Tests: Frontend 28/28; LLM Engine 117 passed, 2 failed; DB validator 112 OK, 0 errors; simulated P95 24.58 ms.
11. Results – ML: ML01 R²=0.873; ML09 R²=0.8165; ML16 rejected; ML19/ML20 proxy (no real data). Table: validated / proxy / rejected.
12. Discussion: Strengths: modularity, security, honest evaluation. Limits: no live validation with real PostgreSQL/Gemini/ABS data, 14 data contracts missing, no push alerts.
13. Demo: screenshots placeholder (landing, AI Lab, forms).
14. Conclusion: functional, statically validated. Next: real data, live Gemini, notifications. Skills: PostgreSQL, MQTT, FastAPI, React, critical ML evaluation.
15. Questions
Created using ChatSlide
This project focuses on developing Agentic AI for industrial supervision within the ABS computer and textile production context. It addresses key problems and objectives while adhering to engineering constraints. Utilizing an Industry 4.0 approach, the architecture integrates IoT, machine learning, and RAG agents, with a data pipeline involving RFID, MQTT, and PostgreSQL. Validation tests show strong performance, with ML results indicating high accuracy. The project outlines lessons learned,...