In the Asah Program led by Dicoding in association with Accenture, my team and I developed a solution aligned with a real enterprise challenge. We selected the Accenture use case because it reflected actual industry demands: operational efficiency, predictive intelligence, and scalable AI deployment. Our solution, AEGIS (AI Engine for Grounded Inspection Systems), was built as an AI-powered predictive maintenance copilot and ultimately received the Best Capstone Project award.

AEGIS addresses a critical industrial problem: reactive maintenance. In many operational environments, machines are repaired after failure occurs, leading to downtime, financial losses, and safety risks. Our approach shifted the paradigm from reactive to predictive. By leveraging machine learning models for failure detection and Remaining Useful Life forecasting, combined with Gemini 2.5 Pro as a conversational intelligence layer, we built a system that does more than predict. It communicates insights clearly to engineers in natural language, including voice-based interaction.

Engineers can interact with AEGIS in everyday language, asking which machines require maintenance, why a certain unit is at risk, or how long before a specific component reaches failure threshold. The system provides grounded, contextual recommendations rather than generic AI responses. This human-centric interface transforms complex predictive analytics into actionable operational decisions.

What AEGIS Delivers 🤖

AEGIS provides a comprehensive predictive maintenance ecosystem, including:

• Operational machine failure detection
• Remaining Useful Life forecasting
• Maintenance recommendations based on risk levels
• Voice command integration
• NLP messaging powered by Gemini 2.5 Pro
• Serverless deployment architecture on Google Cloud Run

The system was designed to be scalable, modular, and enterprise-ready.

My Contributions 🛠️

I contributed in two major capacities: product leadership and technical implementation.

As Project Manager 😎

I led the development using agile methodologies with a strong focus on MVP delivery. My responsibilities included:

• Defining the product roadmap
• Translating business requirements into technical execution
• Leading sprint planning and coordination
• Managing engineering collaboration
• Prioritizing high-impact features

The strategy was execution-driven. We focused on building a working predictive core first, then layering the AI copilot and voice capabilities. This disciplined approach ensured we delivered a functional and scalable solution within the program timeline.

As Machine Learning Engineer 🧠

Beyond leadership, I architected and built the AI copilot layer. This involved:

• Designing the Gemini 2.5 Pro LLM integration
• Engineering the full NLP messaging pipeline
• Creating a grounded response mechanism tied to operational data
• Developing the serverless backend on GCP Cloud Run
• Structuring the full end-to-end AI inference workflow

The result was not just a predictive model but an intelligent assistant capable of contextual reasoning over machine data.

Why AEGIS Stood Out 🤯

Several factors differentiated AEGIS from typical capstone projects:

• Strong alignment with an enterprise-grade use case
• Integration of predictive ML with LLM-based conversational AI
• Cloud-native, serverless architecture
• Clear business impact narrative
• Practical usability for engineers

We approached the project not as an academic exercise but as a real digital transformation initiative.

Key Takeaways 🎯

This project reinforced several strategic insights:

• Predictive models gain exponential value when combined with LLM copilots.
• AI systems must prioritize usability, not just accuracy.
• Cloud-native architecture accelerates deployment and scalability.
• Strong product leadership is critical for successful AI delivery.

Winning the Best Capstone Project award validated both our technical depth and our product-driven mindset. AEGIS demonstrates how AI, when architected strategically, can transform industrial operations into intelligent, proactive systems.