Marco Berta

Designing and industrializing an enterprise AI platform and agent ecosystem, moving GenAI solutions from POCs to production. Key areas: - Architecting RAG and agentic AI systems using MCP/agent servers and enterprise data sources. - Designing integrations between AI agents, enterprise systems and Databricks. - Implementing AI observability, monitoring and FinOps for production systems. - Establishing patterns for scalable, maintainable and governed GenAI applications. Stack : Python, Azure DevOPS, DataBricks, Jira, MCP, AI agents, GenAI/LLM.

Built and industrialized a production RAG system for technical documentation covering PLCs, functional specifications, process diagrams and equipment manuals. Impact: reduced information retrieval from 15–30 minutes to ~6 seconds, with 95%+ relevant retrieval results. Scale & engineering: 10,786 semantically chunked documents, sub-second hybrid retrieval, incremental indexing, metadata-aware retrieval and sourced answers. Stack: Python, LlamaIndex, Qdrant, Azure OpenAI, Docling, Ollama embeddings, Chainlit.

As a content creator and video editor for the Soul Surf Project, I was responsible for creating engaging and informative video content that highlights the project's mission and impact.

Built backend components for an AI-assisted complaint-handling system for the European Commission/DG-GROW. Developed a RAG + LLM workflow that retrieved similar historical complaints and generated draft responses with ~90% reduction compared to manual search. Stack : VSCode, Jenkins, Sonarqube, Azure SQL DB, pytest, Langchain, FastAPI.

- Failure prediction model for wind turbines deployed to the cloud with CI/CD pipelines (Azure DevOps). 0.4% increase of annual energy production at park level. - Time series analysis AI model for signal anomaly detection running on Azure ML pipeline: ⋍20/k/month savings estimated per production tool. Stack : VSCode, Git, Sonarqube, Azure SQL DB, pytest, Langchain, Azure Foundry.

- Built an AI app to suggest the best settings for critical process parameters to maximize cocoa production. The improvement from this advisory tool saved 100k/month per production line. - Collaboration with data engineers to build ML/AI pipelines on local servers and on MS Azure cloud. Real-time dashboards deployed to production lines saved ⋍400/k/year improving dosing accuracy and decreasing downtime. Stack : VSCode, Dataricks, Azure Devops, Kanban, MLOps, Streamlit.

• Experimental data analysis using Python: 95% time saving compared to dedicated scientific software. Data processing automation saved 10-15k for a R&D project budget. • Managed an R&D project work package ,30K semester budget, targeted to improve automotive materials durability. The improvement reached 25% with <1% material cost increase.

• Collected and modeled experimental data (linear regression, exponential fitting). Vegan meat substitute processing optimized according to the model lead to the a product 20% cheaper than beef with 90% reduced CO2 footprint. • Image analysis of liquid foams using ImageJ, correlated to mechanics, produced more accurate measurements of foam stability than that of the traditional foam height test (2% vs. 10%standard deviation).

Managing a project targeted to industrial customers - microelectronics industry (200K annual budget). The resins developed for 3D-chips technology could be precisely machined by laser within less than a second, with 10% processing cost than traditional VIA chips.

- Managed academic research project funded by 100k Marie-Curie grant. Results were published in high impact peer-reviewed academic journals.

Management of a project (100k budget) for the development of novel formulations for personal care products. Results helped to develop skin-repairing cosmetics. Some results obtained within the academic collaboration with Cambridge University were published in peer-reviewed international scientific journals.

- microscopy - corrosion assessment - communicating and discussing results with customers