Back to portfolio

Jul 2025 – Present

Merchant AI assistant

Merchants ask for reports, analytics, and support in plain language — and get the answer and the action instantly, instead of waiting on the data team.

Role: Senior Data Scientist · Zeal (London / remote)
  • Dialogflow CX
  • MCP
  • n8n
  • Langflow
  • Docker
  • GCP
From a merchant's question, the assistant recommends the next action — build a report, surface a KPI, open a ticket.

The problem

At Zeal, every report, analytics question, and support ticket routed through the data and support team. Merchants waited hours — sometimes days — for a single number, and the team drowned in repetitive, near-identical requests instead of doing high-value work. The bottleneck was human, and the need was clear: let merchants self-serve in natural language.

The approach

  1. 01

    Conversational assistant on Dialogflow CX

    I built a conversational interface on Dialogflow CX that merchants talk to in plain language — asking for a report, a metric, or opening a support ticket — with no menus, forms, or waiting on anyone.

  2. 02

    Wired to systems via the MCP concept

    Using the Model Context Protocol (MCP), the assistant reaches data, reports, and support systems through one consistent layer — so it doesn't just answer, it performs a real action on the merchant's behalf.

  3. 03

    Orchestrated AI workflows

    I designed and orchestrated the supporting workflows across n8n, Langflow, and Dialogflow, so a conversational request becomes a chain of steps — fetch the data, assemble the report, file the ticket — that runs end to end, automatically.

  4. 04

    Shipped on Docker Compose, fully tested

    The system ships entirely on Docker Compose with full test coverage — every path is covered, deployable, and reproducible, rather than fragile code that works on luck.

The result

  • Full coverage tests across every assistant path
  • Self-serve reports, analytics & tickets in natural language
  • ~instant answers vs. waiting on the data team (illustrative)
  • ~ requests deflected repetitive asks handled automatically (illustrative)

This is what it looks like when AI doesn't stop at the demo: a reliable conversational assistant wired into real systems, taking real actions, and freeing an entire team from repetitive work to focus on what actually matters.

Next project Forecasting & segmentation Back to portfolio