Ahmed Shawky — Lead Expert Scientist, Data & AI

Ask my AI

An assistant trained on my career, projects, and talks. Ask it anything you'd ask me in an interview — it will tell you what it doesn't know.

Who I am

I'm part of Arelion's AI Center of Excellence, where I build and run production ML and agentic AI systems across network capacity, automation, and data architecture. Before that I spent fifteen years in telecom networks — planning, capacity, and financial modeling — which is why the systems I build answer the questions planners and executives actually ask.

PhD in Telecommunications Engineering from Aalborg University. Copenhagen since 2007, Cairo before that. English, Arabic, Danish.

Oh, and I'm Batman.

Production ML & forecasting
Agentic AI, MCP, Claude Code
GraphRAG & NL→SQL
n8n automation
Python, R, SQL, Snowflake
Telecom infrastructure strategy

Recent work

Lead Expert Scientist, Data & AI

Arelion, Copenhagen — since August 2026

Shipped a production ML system that forecasts network capacity breaches, giving planners a ranked view of where the backbone runs out of headroom before it happens.
Designed a GraphRAG retrieval architecture over 50+ Salesforce objects in Snowflake, so commercial teams ask questions in plain language instead of waiting on a dashboard.
Built a custom MCP server connecting Claude Code to Microsoft Copilot Studio — internal AI agents developed and maintained as code.
Automated network route planning with n8n, replacing manual steps that previously ran on tribal knowledge.

Lead Strategic Data Scientist

Arelion — June 2025 to July 2026

Owned forecasting models for power, site, lease, and maintenance spend that guide pricing and capital strategy.
Mapped physical and virtual network assets to product outcomes so leadership can prioritize investment.
Delivered executive dashboards for the VP of Network Technology & Customer Operations and CEO.
Built reproducible analytics pipelines with R, Python, Airflow, Docker, and SQL.

Earlier: Arelion, Akamai, TDC NET, Syntropy, Aalborg University

Senior Network Data Scientist and Network Financial Analyst at Arelion (2018–2025): built and maintained the Long Run Incremental Cost model, mapped network inventory to sellable services, tracked installed vs. unutilized capacity.
Led EMEA capacity planning at Akamai, combining time-series modeling (Prophet, ARIMA, Holt-Winters) with partner negotiations.
Designed IP network upgrades and business cases as a Network Planning Engineer at TDC NET.
Supported CDN and infrastructure initiatives for Syntropy and guided telecom research at Aalborg University.

What I build

Things I've shipped with AI — at work and on my own time.

Salesforce Semantic Graph

Arelion — NL→SQL assistant

Schema graph of 50+ Salesforce objects (Neo4j metadata, Snowflake data) grounding an LLM that answers commercial questions in plain language, via Slack.
Golden eval set; about 60% of queries route to deterministic metric SQL with no LLM in the loop.

Text-to-SQL: ground it or teach it?

Arelion — three-arm experiment

Graph-grounded vs. static-schema vs. a fine-tuned Gemma 4 (LoRA on Apple Silicon / MLX), scored on one held-out set for accuracy, latency, and cost.

Claude Code ↔ Copilot Studio MCP server

Arelion — agents as code

Custom MCP server so internal AI agents are developed, versioned, and maintained in a repo rather than clicked together.
Shared Claude Code skills for branded reporting and querying the network and commercial semantic model.

Nexus & TETRIS

Arelion — optical network intelligence

Nexus: WDM capacity visualization and route planning — path finding, transponder availability — Python and Docker, in production.
TETRIS: finds sellable free spectrum per WDM system, maps gaps to transponder cards, answers end-to-end route feasibility.

n8n automations

Arelion — workflow automation

IP route planning, port cabling, and RFQ tracking workflows, built and maintained through n8n's MCP server from Claude Code.

This site's assistant

Personal — Gradio and OpenAI tool calling

Answers questions about me, records leads and unanswered questions. Hosted on Hugging Face Spaces, fed from the same profile file as this page.

Crypto trading agent

Personal — Claude-driven, code-enforced risk

Daily systematic agent on OKX demo: candles and news in, one structured Claude decision, hard risk rules, exchange-native stops, a journal entry.

Faraho

Personal — bilingual marketplace

Egypt wedding-venue marketplace in English and Arabic with full RTL: React 19, Firebase, Turborepo, Playwright e2e, deployed on Vercel.

Talks & podcasts

Memory has no changelog

Speaker, CPH^Automate — Copenhagen, 2 October 2026

We asked two route planners to write down how they route fibre and got back 98 sections — which described routing by who owns the building, while the same engineers had for years actually been routing by what's physically installed there. How we stopped asking people to remember, recovered the real rules by replaying historical decisions out of the inventory database, and the one line of code that got one in four trunk selections wrong without ever raising an error.

Event and speakers

Evolution Exchange Denmark

Podcast panels — Using AI Effectively (August 2026) and The Growth of AI in Europe (March 2026)

Education

PhD, Telecommunications Engineering — Aalborg University, 2016
MSc, Networks and Distributed Systems — Aalborg University, 2009
BSc, Computer Engineering — October 6 University, 2006
Claude Platform 101 — Anthropic, 2026. Four IBM data science certificates, six DataCamp R and Python tracks.

Ahmed
Shawky