Piktechs
B2B SaaS platform for digital business cards at professional events: connection, lead capture with an AI card scanner, CRM and team workspaces. Designed, built and operated end to end.
Founder & Full-Stack Engineer · Software, SaaS & AI
I ship software & AI products to production.
About
I build software and AI systems end-to-end: specs, architecture, code, deploy. Shipping production software since 2019 across cybersecurity, healthcare and fintech, often as the only engineer on the project. I work with founders and agencies who need something built properly and shipped fast.
What I do
Your product built end-to-end: specs, architecture, code, deploy. From idea to a production v1 your users can actually use. I handle the whole stack.
For founders who need a v1 shipped properly
LLM-powered workflows, AI agents, RAG and OCR/Vision pipelines that remove manual work from your operations, built on your real processes, not demos.
For teams drowning in repetitive work
Chat over your documents, OCR + LLM pipelines, RAG with memory, OpenAI / Claude / Gemini inside your existing product — with rate limits, fallbacks and evaluation handled.
For products that need AI features that hold in production
Projects
B2B SaaS platform for digital business cards at professional events: connection, lead capture with an AI card scanner, CRM and team workspaces. Designed, built and operated end to end.
SaaS converting PDFs to structured Markdown through a multi-pass OCR pipeline combined with Gemini. Understands and transforms documents of any size, in 11 languages.
Self-service tool measuring how AI answer engines describe a brand across ChatGPT, Perplexity, Gemini and Google AI Overviews. Statistical sampling per prompt, real geo-localization and week-over-week tracking.
Online store of Montessori educational toys and games, built end to end, grown to real revenue, and successfully sold: SEO, marketing and custom Liquid themes.
They trust me
Learning how to use AI effectively is becoming a real lever for efficiency and impact. During our ChatGPT Workshop, "Make AI work for you", we explored how AI can support us from the very first question to the automation of a complete business process. A big thank you to El Mahdi MOUBARAK for leading this insightful and energizing session. He brought both technical depth and a very practical mindset.
Chief People Officer · DRIVECO
Read the post on LinkedIn ↗
“We trusted El Mahdi with the mobile app for our delivery service. He taught himself what he needed and built the app autonomously. I recommend him strongly.”
CEO · Deuspi
As CTO of AGENCE-KSUR, I own technical delivery for the agency's clients: websites, e-commerce and web apps shipped for SMEs in France and Morocco.
See the agency portfolioCompanies I've built for
Writing
I have about ninety Claude Code skills and I don't remember their names. Replaying my own prompts showed Claude loaded only 57% of the skills a request needed. A hook in my own words took it to 84% on prompts it was never tuned on.
Same mini PC, same cable, an RTX 3080 with 20 GB: Qwen3.8-Flash-Next 180B now reads about 2,000 tokens per second and writes 110 to 204. My settings and benchmarks.
Qwen3.8-Flash-Next, 180B parameters, running with Strata on a mini PC and an RTX 3060 Ti 8 GB over OCuLink: about 390 tokens per second reading, 55 writing. The numbers, and how they got there.
Contact
Open to interesting projects and conversations.