Siskeu BUMDes AI Operator

AI-operated financial assistant for BUMDes, powered by Gemini and LynkMesh to explain reports, check data, and support human-approved workflows.

Comment

Inspiration

Many village-owned enterprises in Indonesia, known as BUMDes, need to manage financial administration, reports, and accountability with limited accounting support. Software can help, but many users still need guidance to understand reports, check incomplete records, and operate financial workflows correctly.

Siskeu BUMDes was created to support this need. It has been used in a real BUMDes context, and the original application development was paid for by a real customer. That historical payment shows validated demand, not just a technical experiment.

For this hackathon, we are building a new AI-operated service layer on top of that foundation: Siskeu BUMDes AI Operator.

What it does

Siskeu BUMDes AI Operator helps BUMDes users understand and operate financial administration workflows with AI assistance.

The first workflow focuses on monthly financial reporting:

  • explaining financial reports in plain language,
  • identifying missing or incomplete data before closing,
  • generating AI-assisted report briefings,
  • keeping a timestamped evidence log of every AI run,
  • preserving human review for sensitive financial decisions.

The goal is not to replace human responsibility. The AI explains, checks, and proposes. Humans remain responsible for approval and final decisions.

How we built it

The project has three layers:

  1. Siskeu BUMDes — the existing financial administration platform and real-world domain foundation.
  2. LynkMesh — the context infrastructure that maps the application structure and produces compact AI-ready context.
  3. Siskeu BUMDes AI Operator — the new Gemini-powered service layer built for this hackathon.

The AI Operator uses Gemini for report explanation and operational assistance. It uses LynkMesh context so the model is grounded in the actual software structure instead of acting like a generic chatbot.

The planned deployment uses Google Cloud Run for the service, Gemini API for AI workflows, Cloud Logging for execution records, and a simple evidence dashboard to show agent runs, token usage, context IDs, and human review status.

Challenges we ran into

The biggest challenge is building AI assistance that is useful without overclaiming autonomy. Financial administration needs trust, traceability, and human oversight.

Another challenge is separating historical traction from new hackathon revenue. The original Siskeu BUMDes application has historical paid development revenue, but the AI Operator layer will be packaged as a new paid pilot/add-on during the hackathon.

We also need to protect sensitive financial data. For demos and public screenshots, the system should use synthetic or redacted tenant data only.

Accomplishments that we're proud of

We are proud that this project is built on a real financial software foundation, not a toy example. Siskeu BUMDes has real-world usage and historical paid development revenue.

We are also proud of the LynkMesh integration concept. Instead of sending a large codebase directly into an AI model, LynkMesh produces compact context that helps Gemini understand the relevant application structure.

The evidence pipeline is another key accomplishment. Every AI run is designed to be logged with a timestamp, LynkMesh context ID, Gemini model, token usage, prompt version, output status, and human review status.

What we learned

We learned that AI for financial administration must be designed differently from a general chatbot. It needs boundaries, logs, review steps, and clear evidence.

We also learned that the strongest AI-native businesses are not only about model output. They need a full operating loop: real user problem, real workflow, AI assistance, human approval, deployment logs, and revenue evidence.

LynkMesh helped clarify an important lesson: context is the moat. A good AI operator needs structured context from the real system it supports.

What's next for Siskeu BUMDes AI Operator

The next step is to launch the first AI-assisted pilot: AI-Assisted Monthly Closing & Report Briefing for BUMDes users.

We will focus on:

  • deploying the first Gemini-powered report explanation workflow,
  • logging every AI action,
  • preparing a simple evidence dashboard,
  • collecting historical traction evidence,
  • generating new AI Operator revenue through a paid pilot,
  • expanding into pre-closing checks, support assistance, and human-approved financial workflows.

Our goal is to help BUMDes operate financial administration more clearly, safely, and efficiently with AI.

Built With

Share this project:

Updates