Skip to content

GMC For-Benefit Economy RAG Agent

  • :material-brain:{ .lg .middle } Reasoning Engine

    7 structured reasoning paths aligned to GMC's for-benefit philosophy. Auto-detects which paths apply to any economic proposal.

  • :material-file-document-multiple:{ .lg .middle } Document-Aware

    58+ catalogued documents from gmc.bt, Fourth Sector Group, Wikipedia, and media. All embedded and retrievable with hybrid search.

  • :material-account-group:{ .lg .middle } Team Partner

    Session workspaces, decision logging, gap analysis, and comparison mode. Not a decision-maker — a reasoning partner.

  • :material-open-source-initiative:{ .lg .middle } 100% Local & Open

    Ollama + ChromaDB + open-source embeddings. No API keys. Runs entirely on your machine. English + Dzongkha capable.

Document Description
Product Requirements Document Vision, personas, features, roadmap, success metrics
System Prompt & Reasoning Paths Full system prompt with all 7 reasoning path templates
Information Directory Catalog of 58 document sources with ingestion priority
Technical Architecture Stack, data flow, component diagrams, deployment

The 7 Reasoning Paths

# Path What It Evaluates
1 Pillar Alignment Alignment with GMC's 8 core industries
2 For-Benefit Design Social Purpose, Earned Income, and 8 secondary characteristics
3 GNH Impact Effects across 4 pillars and 9 domains of Gross National Happiness
4 Sustainability Zero-waste, carbon-negative, renewable, biodiversity compliance
5 Legal / Regulatory SAR Basic Law, licensing, tax, digital asset framework
6 Ecosystem Mapping 10 Fourth Sector ecosystem elements; existing vs. needed
7 Mindful Capitalism Balance of prosperity and mindfulness; youth, culture, community

Technical Stack

┌──────────┐    ┌──────────┐    ┌──────────┐    ┌──────────┐
│  Ollama  │    │ ChromaDB │    │  SQLite  │    │  Python  │
│  Qwen 2.5│    │  Vector  │    │  FTS5    │    │  Agent   │
│  (LLM)   │    │  Store   │    │  BM25    │    │  Runtime │
└──────────┘    └──────────┘    └──────────┘    └──────────┘

Embeddings: intfloat/multilingual-e5-large (768-dim, multilingual) Reranker: BAAI/bge-reranker-v2-m3 (cross-encoder) LLM: Qwen 2.5 7B/14B via Ollama (128K context, Dzongkha capable) Vector store: ChromaDB (persistent, metadata-filtered) Hybrid search: Dense (ChromaDB) + Sparse (SQLite FTS5) fused via RRF

Project Status

  • :material-check-circle:{ .green } Phase 0: Document corpus catalogued (58 sources)
  • :material-circle-half-full:{ .orange } Phase 0: Documents indexed, pipeline designed
  • :material-clock-outline:{ .gray } Phase 1: Core RAG pipeline (next)
  • :material-clock-outline:{ .gray } Phase 2: Reasoning path engine
  • :material-clock-outline:{ .gray } Phase 3: Agent workflows & team features

Last updated: 2026-06-23