ASH Mineral Tech

AI-native industrial framework automating mineral geopolymer engineering, B2B quarry-to-mill operations, and material synthesis workflows via multimodal Gemini agents

Comment

Inspiration

Heavy mineral beneficiation and high-purity industrial silica processing operate in environments plagued by siloed telemetry, manual quality control, and disconnected B2B logistics. ASH Mineral Tech was conceived to bridge deep industrial engineering with an autonomous, 3-tier hierarchical multi-agent framework powered natively by Google's Gemini models—transforming raw quarry-to-mill operations into an intelligent, self-governing industrial ecosystem

What it does

ASH Mineral Tech is a comprehensive AI Operating System operating on four pillars: Factory Telemetry, Fleet Logistics, B2B Commercial Matchmaking, and Financial DRE Governance

  • 3-Tier Multi-Agent Hierarchy:
    • Operational Specialists (Tier 1): Gemini 2.0 Flash agents executing real-time document OCR, equipment checklists, and operator telemetry ingestion
    • Domain Supervisors (Tier 2): Specialized synthetic agents supervising Labor & Compliance (Dra. Cecília), Fleet & Maintenance (Eng. Gustavo), and Finance & Margins (Auditor Marcelo)
    • Executive Orchestrator (Tier 3 - Cássio): High-reasoning executive agent evaluating cross-domain operational friction, projecting DRE impacts, and dispatching authoritative business decisions
  • Multimodal Mesh Inspection: Gemini Vision pipelines analyzing microscopic mineral samples to verify mesh consistency (#100, #200, #325 Mesh) and flag impurities before batch dispatch
  • SPARK Autonomous Gateway: Event-driven REST & telemetry gateway syncing real-time mill metrics and automated production orders directly with cloud inventory
  • Interactive Operations (Charles): Conversational AI interface supporting plant floor operators with technical procedures and regulatory standards

How we built it

  • AI Core: Integrated @google/genai SDK using Gemini 2.0 Flash with strict JSON responseSchema validation for deterministic reasoning
  • Backend & Telemetry: Node.js, Express, and Vite middleware architecture featuring isolated REST endpoints for edge sensor data ingestion
  • Frontend & UX: React 18, TypeScript, Tailwind CSS, Lucide icons, and Motion for reactive plant monitoring interfaces
  • Enterprise Security & Dual-Mode: Engineered an isolated dual-mode environment (VITE_DEMO_MODE=true) running high-fidelity synthetic plant datasets without exposing private keys or proprietary infrastructure

Challenges we ran into

  • Implementing strict JSON schemas for deep reasoning tasks to eliminate hallucinations in mission-critical industrial decisions
  • Handling asynchronous edge telemetry while keeping agent orchestration latencies sub-second for fast-paced logistics and milling workflows
  • Architecting a zero-leak public demonstration mode preserving full multi-agent functionality

Accomplishments that we're proud of

  • Bridging physical mineral processing machinery with next-generation autonomous AI agents
  • Building a full-stack, end-to-end industrial system capable of multimodal vision auditing and autonomous operational governance
  • Ensuring enterprise-grade separation of secrets while delivering instant one-click demo deployment

What we learned

  • Hierarchical agent delegation combined with native Gemini structured outputs provides industrial-grade reliability and auditability that single-prompt architectures cannot match

What's next for ASH Mineral Tech

  • Deploying real-time edge micro-controllers running on-premise vision inference connected to the SPARK Gateway
  • Expanding continuous B2B marketplace matchmaking across regional industrial hubs

Built With

Share this project:

Updates