Zync — Malaysia's Sovereign REE Engineering Intelligence

GLM-5.1's scientific reasoning at the point where Malaysia's RM 1 trillion rare earth ambition is won or lost.

Malaysia's REE ambition depends on solving hydrometallurgical decisions the country doesn't have enough engineers to make correctly at scale. Zync puts sovereign AI reasoning at those three decision points.

The Three Decisions

#	Decision	What it solves	Agent
1	Process Diagnosis	Why did my yield drop?	Agent 2 (streaming SciGLM)
2	Lixiviant Selection	What chemistry for this soil?	Agent 2 + function calling
3	Zone Prioritisation	Which zone do I invest in first?	Agent 6 (scored ranking)

GLM Capability Layers Demonstrated

• Streaming reasoning trace — GLM's thinking visible token by token (Decision 1 & 2)
• Function calling — lixiviant knowledge base lookup mid-reasoning (Decision 2)
• Long context multi-variable scoring — all zone profiles assessed simultaneously (Decision 3)
• Bahasa Malaysia output — compliance and zone recommendations in BM (Decision 3)
• Known answer validation — 5 tests against published Malaysian REE literature

Tech Stack

• AI — Z.ai GLM-5.1 (SciGLM, 202K context, streaming reasoning)
• Backend — FastAPI (Python), Server-Sent Events for streaming
• RAG — GraphRAG (Neo4j), Hybrid Search (fastembed + Qdrant)
• Frontend — React + Vite, Bahasa Malaysia + English

API Endpoints

Endpoint	Description
POST /api/pipeline	Full 6-agent SSE pipeline (Decisions 1 & 2)
POST /api/zone	Zone Prioritisation SSE stream (Decision 3)
POST /api/validate	5 known answer tests against Malaysian REE literature
POST /api/upload/pdf	Geological survey PDF extraction
POST /api/admin/seed	Seed Neo4j with Malaysian REE case history

Agent Pipeline

Decision 1 & 2 — POST /api/pipeline
    → Agent 0 (Router)
    → Agent 1 (Historian — GraphRAG)
    → Agent 2 (Chemist — SciGLM ★ streaming reasoning)
    → Agent 3 (Optimizer — GLM iteration loop ★ live streaming)
    → Agent 4 (Compliance — bilingual AELB + DOE + JMG)
    → Agent 5 (Report Writer)

Decision 3 — POST /api/zone
    → Agent 6 (Zone Prioritiser — 5-step scoring ★ streaming)

Quick Start

1. Configure environment

# Add your GLM_API_KEY to backend/.env
# (copy backend/.env.example if it doesn't exist)

2. Start frontend + backend + all databases (Docker)

docker compose up --build
# Backend API → http://localhost:8000
# Neo4j, PostgreSQL, Qdrant all start automatically

3. Seed the databases (first run only)

curl -X POST http://localhost:8000/api/admin/seed -H "X-Admin-Secret: zync-admin"
curl -X POST http://localhost:8000/api/admin/seed-qdrant -H "X-Admin-Secret: zync-admin"

The frontend proxies /api/* to the backend at http://localhost:8000 (see frontend/vite.config.js).

Daily startup: docker compose up (no --build needed unless code changed). Stop with docker compose down.

Team

Built for UM Hackathon 2026.