Ariel Rubinstein

First fruit.

Featured builds

Operational systems I built across finance, airline analytics, AI decision support, e-commerce, and customer service. I own the architecture, implementation, and the details that make them useful in daily work.

• Cash-flow and reconciliation

  Operational finance system · air-cargo GSA

  Built a cash-flow application that turns bank statements and spreadsheet exports into a monthly management dashboard. Python workers identify and reconcile transactions across companies and four currencies; every total drills down to its source records. Bookkeepers review exceptions while existing human decisions stay authoritative.

  • Source-linked transactions and management reporting
  • Multi-company, four-currency reconciliation

• Airline BSP analytics

  Operational analytics platform · airline client

  Built billing-archive parsers and a dashboard for sales, refunds, and debit/credit memos. Agency and branch drill-downs, month reviews, settlement reporting, and audited exports turn source reports into actionable analysis. Reconciliation checks and explicit coverage windows keep partial periods from becoming misleading comparisons.

  • From billing archives to management analysis
  • Source-reconciled totals and memo drill-downs

• Jev decision engine

  Operational AI system · financial classification

  Implemented hierarchical transaction classification with Jev, including confidence, abstention, and traceable decisions. Built held-out bookkeeping evaluations and a Claude coaching loop that proposes and tests improvements. Human corrections, model suggestions, and authority to assign automatically are handled separately.

  • Evaluation and coaching built into the workflow
  • Human review and controlled assignment authority

• Instagram and customer support

  Operational automation · e-commerce

  Built Instagram and Facebook workflows for publishing approved content, answering knowledge-base questions, capturing purchase intent, and handing customer conversations to a human. n8n, Meta APIs, and a shared conversation store connect publishing and support, with approval controls, duplicate-message handling, and operational alerts.

  • Approved content to scheduled publishing
  • Knowledge-base replies with human escalation

• Tabletops by Nomi

  Operational e-commerce platform · client build

  Designed and built a bespoke table-linen e-commerce platform with a custom-dimensions pricing configurator, checkout, and internal admin and inventory tools. Next.js, Supabase, and Stripe connect to 19 n8n workflows for shipping, invoicing, payment reconciliation, and business operations.

  • Architecture through deployment and operations
  • 19 workflows connect storefront and back office

• Document-intelligence stack

  Operational knowledge infrastructure · self-hosted

  Kreuzberg text extraction feeds a LightRAG knowledge graph with hybrid semantic search. An MCP server exposes four knowledge-base tools to Claude Code agents, grounding automation in accumulated context. A multilingual RAG and NotebookLM sync system spans 15 knowledge bases and 83 sources.

  • Knowledge retrieval available directly to agents
  • Scheduled ingest and knowledge-base synchronization