SafeServe MBG

The unified AI command center for the Makan Bergizi Gratis program. Real-time oversight for 83 million beneficiaries.

Comment

About SafeServe MBG

Inspiration

"No child should ever have to trade their health for a meal."

The inspiration for SafeServe MBG was born from a place of urgency and deep concern. As Indonesia embarks on the historic Makan Bergizi Gratis (MBG) program to nourish 83 million children, a darker reality has emerged: headline after headline reporting hundreds of students hospitalized due to food poisoning in regions like Tasikmalaya, Cianjur, and beyond.

While the program is a beacon of hope for building the "Golden Generation 2045," the current reliance on manual, reactive supervision has proven fatal. We realized that for a parent, the joy of a free meal is quickly replaced by the terror of a phone call from the school clinic. We built SafeServe to be the Digital Shield that was missing—a system that moves beyond paperwork to provide real-time, uncompromising oversight. We believe that "Free" must never be an excuse for "Unsafe," and that technology is the only way to ensure every lunchbox served is a promise of health, not a hidden hazard.

What it does

SafeServe MBG is a unified intelligence platform that connects the entire food safety loop:

  • For Regulators: Provides a national "Command Center" with AI-driven outbreak prediction, regional risk heatmaps, and deep "Reasoning Chains" that explain the logic behind vendor risk scores.
  • For Vendors: Offers a portal for digital transformation, including AI-powered multimodal hygiene audits (vision), staff wellness tracking, and blockchain-verified supply chain traceability.
  • For Schools: Empowers admins with SOS symptom reporting (linked to AI clinical triage), shipment verification, and nutritional compliance monitoring.
  • For Sustainability Officers: Implements an "Active Governance" loop to reduce food waste through AI-suggested portion calibrations and district-level consumption analysis.

How we built it

The platform is built on a high-performance frontend stack designed for mission-critical reliability:

  • Core: React 19 and TypeScript for a robust, type-safe architecture.
  • Intelligence: Integrated the Google Gemini API (gemini-3-flash-preview and gemini-3-pro-preview) for multimodal vision analysis, text-based triage, and complex reasoning.
  • Grounding: Utilized Google Search Grounding to inject real-time regional news and stability data into our risk models.
  • UI/UX: Tailored with Tailwind CSS 4 and Lucide React, focusing on a "Command Center" aesthetic that remains accessible for local school staff and UMKM owners.
  • Data Viz: Leveraged Recharts for real-time telemetry from IoT cold-chain simulators and waste benchmarking.

Challenges we ran into

  • AI Explainability: One of the biggest hurdles was ensuring the AI didn't just give a "Risk Score" but could explain why in a way a regulator could trust. We solved this using Gemini's reasoning capabilities and structured output.
  • Multimodal Integration: Designing a system that could handle both real-time IoT sensor data and image-based hygiene audits required careful state management and resilient API handling (implementing circuit breakers for quota management).
  • User Diversity: Designing a single ecosystem that caters to high-level government officials and hyper-local catering staff required a deep focus on intuitive UX and clear action-oriented design.

Accomplishments that we're proud of

  • Unified Safety Loop: Successfully creating a "Single Pane of Glass" where a report at a school in Jakarta can immediately update the risk profile of a vendor and alert a national regulator.
  • AI Triage Accuracy: Developing a clinical triage system that maps reported symptoms to potential pathogens and provides immediate safety protocols.
  • Simulation Mode: Building a fully functional "Outbreak Simulator" that allows officials to test their response systems in a "war-room" style environment.

What we learned

Building SafeServe MBG taught us the immense value of Grounding AI. Without real-world context (via Google Search), a safety model is just a calculator. With it, it becomes an intelligent partner. We also learned that in mission-critical applications, "Trust" is the most important feature—which is why every AI decision in SafeServe is backed by evidence chains and integrity hashes.

What's next for SafeServe MBG

  • Real-time Video Audits: Moving beyond static photos to live-streaming kitchen analysis using Gemini's Video capabilities.
  • IoT Hardware Integration: Moving from simulated telemetry to real-world sensors in delivery trucks.
  • Local Language Support: Enhancing the AI Assistant to support regional Indonesian dialects for broader inclusivity among UMKM vendors.
  • Predictive Procurement: Using consumption data to help vendors pre-order ingredients, further reducing waste and costs.

Built With

Share this project:

Updates