Participating in the Amartha x GDG Jakarta Hackathon 2025 was a fast, high-stakes learning curve. Our team, Extra Pika Pika, went in with zero expectations. It was our first hackathon, our first large-scale collaborative build, and our first attempt to convert a conceptual risk framework into a production-grade Early Warning System.

We made it to the Top 15 finalists. We didn’t win, but the experience delivered a level of operational insight that no online course or tutorial can replicate. The environment was intense, timelines were narrow, and expectations were real. It pushed us into a mindset of execution over theory and alignment over ego.

The Core Problem We Tackled 🧠

Financial risk in microfinance is structurally complex. It is not just about individual borrowers. It is about groups and the agents who manage them. This was also the core challenge highlighted in the official Amartha playbook, which became the foundation for our solution concept.

We realized two gaps:

1. Group behavior has a contagion effect. One default can escalate across the entire cluster.

2. Field agents hold critical contextual intelligence, but that information is unstructured and underutilized.

Traditional scoring tools do not capture these dynamics. That is where SATRIA originated.

What We Built? It's Called SATRIA 🤺

SATRIA stands for "Sistem Analisis Terpadu Risiko Agen & Grup" in Indonesian and "System for Analyzing Total Risk of Agent & Group" in English. Both maintain the same abbreviation to ensure consistency across languages. The platform is an AI-driven Early Warning System built to deliver proactive, data-centric risk visibility across Amartha’s operational ecosystem.

The solution had three operational pillars:

1. Agent Pulse Analytics
We used NLP to analyze sentiment and behavioral indicators found inside agent field reports. This allowed us to produce automated agent performance and risk scoring at scale.

2. Group Dynamic Risk Model
We leveraged aggregated repayment metrics as input signals. Instead of over-engineering with GNNs at MVP stage, we used Gemini’s reasoning capability to detect early signs of group fragility and potential contagion scenarios.

3. GenAI Advisor Dashboard
We focused on explainability. The system didn’t just detect anomalies. It recommended interventions. Regional managers could see what was wrong and understand why it mattered.

This directly aligned with the needs of operations, risk teams, and training functions.

How We Executed Under Hackathon Constraints 🛠️

We structured the implementation strategy to maximize output within limited time:

• Backend in Go for efficient orchestration.
• PostgreSQL for structured context storage.
• Prompt engineered workflows to offload complex reasoning to Gemini.
• React dashboard as the single view for agent and group health insights.
• Google Cloud Platform for deployment using Cloud Run.

This approach kept our MVP lean and fully demo-ready.

The Experience Behind the Scenes 🤯

Nothing prepares you for the tempo of a hackathon until you are inside one.

As a first time participant, several things stood out:

Collaboration is a forcing function

Everyone had to operate in lockstep. If the backend was delayed, the frontend stalled. If the prompt engineering was off, the risk scoring broke. It forced us to communicate with clarity and move with shared accountability.

Constraints accelerate innovation

With limited time, we had to be ruthless about scope. That forced smarter design decisions and more pragmatic builds.

Technical alignment beats individual expertise

Every teammate had strong skills. What mattered more was alignment and clarity of roles. That was the real productivity multiplier.

Losing doesn’t remove value

Not winning was fine. The operational maturity we gained was worth more than a trophy.

What I Personally Learned? 🤔

As a data scientist and cloud engineer on the team, this experience reinforced several truths:

• Shipping beats theorizing.
• Finished is better than perfect.
• Real users care about clarity, not complexity.
• Team dynamics are as important as technical design.
• Pressure reveals how you operate when the clock is against you.

Most importantly, I learned how to operate as part of a team that is fully locked in on one shared value proposition.

What’s Next? ✌️

We plan to continue iterating on SATRIA beyond the hackathon. The concept has potential to scale into a fully operational risk intelligence platform, especially with improved data pipelines and advanced modeling.

Hackathons aren’t just competitions. They are accelerators. This one accelerated how I think, build, collaborate, and deliver.