NowShipping agentic analytics

I build AI where the work actually happens.

Forward Deployed Engineer · Middle East

I'm Hicham — a forward deployed engineer. I embed with project, risk and bid teams, find the workflow that hurts, and ship production AI systems: from the SQL warehouse to the last slide of the deck.

Scroll

How I work

The forward deployed loop

Most AI projects stall in the gap between a good demo and a normal Monday. I close that gap by working inside the team until the tool is simply part of how they work.

  1. 01

    Embed

    Sit with the team that owns the problem. Watch the monthly pack get built by hand, and note every place someone copies a number from one window to another.

    A map of the real workflow

  2. 02

    Frame

    Find the one decision the data should inform, and the rules people already apply to it. Those rules become code; only the judgement goes to the model.

    A thin, testable slice

  3. 03

    Build

    Ship against live systems early: real warehouse, real documents, real permissions. Demos on sample data hide exactly the problems that matter.

    A working tool on real data

  4. 04

    Harden

    Access control, audit logs, graceful failure, self-tests, a runbook. The tool has to keep working on the Monday after I’ve moved on.

    A service the team owns

What I bring

The whole path, from warehouse to Monday morning.

  • LLM systems that hold up

    Prompts written as mechanical rules, strict-JSON extraction, streaming protocols, content-filter recovery and model-agnostic clients — so swapping a deployment is a config change.

  • Data plumbing

    SQL Server procedures, a governed knowledge graph, pure-Python aggregation layers and schema-drift-tolerant writes.

  • Document automation

    PowerPoint, PDF and Outlook-safe email generated from structured data — and previewed as the real file.

  • Product surfaces

    React and TypeScript workspaces non-technical teams adopt: live progress, results you can edit in place, dashboards drawn in SVG.

  • Production hygiene

    Single sign-on, per-agent access, audit logs, caches keyed on data versions, and self-tests that need no network.

  • On-prem to edge

    FastAPI behind IIS with rollback by repointing — or static sites deployed to Cloudflare’s edge, like this one.

Tools I use:

  • Python
  • FastAPI
  • Server-Sent Events
  • TypeScript
  • React
  • Azure OpenAI
  • Azure AI Search
  • Whisper
  • SQL Server
  • python-pptx
  • LibreOffice
  • Microsoft Graph
  • IIS
  • Astro
  • Cloudflare Workers

Profile

Engineer by trade. Translator by necessity.

Role
Forward Deployed Engineer
Region
Middle East
Focus
AI for project delivery, risk & bids

I’m Hicham Mokadem, a forward deployed engineer in the Middle East division of a global engineering consultancy. I work where project controls, bids and delivery meet data — and where most of that data still travels by spreadsheet, email thread and slide template.

My job is to sit with the people who run that work, find the decisions that matter, and ship AI systems that make those decisions faster and better grounded. The platform in the case study is what that looks like end to end: discovery, architecture, prompts, plumbing, deployment and hand-over.

I care most about the unglamorous parts: numbers that reconcile, previews that match the file, failures that degrade gracefully, and tools that still work the Monday after launch.

  1. P.01

    Numbers from code, words from the model.

    Arithmetic, dates and provenance are deterministic. The model writes.

  2. P.02

    The preview is the product.

    If a user approves something on screen, the file they download is exactly that.

  3. P.03

    Fail soft, log everything.

    A missing data source becomes a warning, not an outage. Every run leaves a record.

Contact

Have a workflow that should be a product?

Start with the story of one that was.

Read the case study