AI Agents

Four agents, one data platform

Purpose-built AI agents that connect EFI's Airtable CRM and PostgreSQL ERP to deliver market intelligence, trade analytics, sourcing gap analysis, and pricing optimization — all running on a self-hosted TypeScript monorepo with a centralized REST data layer.

Runtime Node.js / TypeScript
AI Models Claude Sonnet 4.6 + Gemini 2.5
Status 4 Agents Live

Architecture

Each agent follows the same pattern: purpose-built fetchers pull data from Airtable and PostgreSQL in parallel, a service layer computes the business logic, and an Express web server renders the results. AI augments the pipeline — Claude writes narratives, validates confidence, and generates recommendations.

AirtableCRM
PostgreSQLERP
AI PipelineWeb Search
DatamyneBOL Data
→
Data API29 endpoints
→
Market Research
Sourcing
Pricing
→
Web Dashboard
PDF Export
Mobile PWA
ⓘ

Self-hosted, not platform-managed. Unlike hosted AI platforms (Google ADK, CrewAI), EFI's agents run on the application server. ERP data never passes through third-party orchestration — only specific prompt data is sent to model APIs. Cost: ~$0.03–0.50 per agent run.

The Four Agents

01 / 04
★

Market Research

The primary EFI application server (port 3000). Hosts the AI market report pipeline (3-pass Claude + Gemini), trade intelligence dashboards (Datamyne BOL data), calculators (Milk ROI, Fat Blending, Product Database), supplier intel email poller, and a mobile PWA. Separate customer insights server on port 3001.

12 data fetchers AI narratives Mobile PWA
→
02 / 04
◆

Sourcing Intelligence

Demand planning with product × month gap matrix, AI-powered sourcing recommendations with capacity scenarios and PDF export, HiGHS LP cost optimizer with 7 what-if scenarios, supplier pipeline and capacity heatmap analytics, forecast audit (11 issue types), and daily digest email.

10 Data API fetchers Rules engine + AI LP optimizer
→
03 / 04
◈

Pricing Optimization

Linear programming model (HiGHS solver) that maximizes margin across the product portfolio subject to supply, demand, inventory, and capacity constraints.

LP/MIP optimizer Scenario comparison What-if analysis
→
04 / 04
◉

Trade Intelligence

Competitive intelligence from US maritime import records. Importer and supplier rankings, port-level trade flows, HS code classification analysis, AI entity resolution and brand discovery, plus calculators (Milk Pricing & ROI, Fat Blending, Product Database). Embedded in the Market Research server at /trade-intel/*.

8 data collectors AI discovery tools 4 calculators
→

Monorepo Structure

directory 
AGENTS_mono/
├── packages/
│   ├── shared/              # @efi/shared — DB clients, models, types
│   ├── market-research/     # Customer insights — fetchers, AI pipeline, web
│   ├── sourcing/            # Gap dashboard — demand/supply/inventory analysis
│   ├── pricing/             # Pricing optimizer — LP model, scenario runner
│   ├── trade-intel/         # Competitive intelligence — Datamyne BOL analysis
│   ├── data-api/            # Centralized REST API — 27 endpoints, 6 route groups
│   └── data-platform/       # This site — static HTML documentation
├── docs/                    # Technical guides and architecture docs
└── .env                     # Shared environment variables

Shared Infrastructure

  • Airtable CRM client — lazy-init, queryTable with formula builder
  • PostgreSQL ERP client — pool with SSL/DigitalOcean compat
  • Anthropic Claude client — singleton, configurable model
  • Data API client — typed REST client for all 28 endpoints
  • Datamyne API client — singleton with 24hr cache, rate-limit protection
  • Product name normalization — fuzzy match across systems

Build System

  • npm workspaces + TypeScript project references
  • tsc -b builds all packages in dependency order
  • Each agent has its own Express server on a separate port
  • Shared .env at monorepo root via dotenv/config

Data Sources

Airtable CRM

TableUsed By
CustomersMarket Research
Sales OrdersMarket Research, Sourcing
Contract Line ItemsMarket Research, Sourcing
Sales Rep ForecastsSourcing
Inventory AvailabilitySourcing
WarehousesSourcing
ProductsMarket Research

PostgreSQL ERP

TableUsed By
supplier_purchase_ordersMarket Research, Sourcing
productsAll agents
ref_sales_monthsMarket Research, Sourcing
ref_otw_transit_timesMarket Research
ref_landed_cost_estimatesMarket Research
ref_efi_portsMarket Research, Sourcing

External APIs

SourceCachePurpose
Datamyne US Maritime Imports (idApp 44)24hBOL records, importer/supplier totals, entity resolution
Datamyne US Import Census (idApp 127)24hFOB pricing by country and month
USDA MPR DataMart (report 2991)30dDairy component prices (butterfat, protein, Class III/IV)