Madrid · Europe · Remote · ES / EN / 中文Taking new engagements

Sami Halawa —Staff AI Systems Engineer & Production AI Architect

I design, ship and operate production AI systems for companies, investors and executive teams.

  • 01Stateful multi-agent systems
  • 02High-throughput search & RAG
  • 03Cloud compute cost reduction
  • 04EU AI Act compliance

Direct B2B contracts, technical due diligence and fractional CTO / CAIO leadership.

Four production pillars

Every system I ship maps to one of four architectures.

Each pillar is backed by systems running for clients, partners or my own products, with the numbers that matter to the people paying for them. Every project is grouped by type on the projects page.

01

Multi-Agent Orchestration & Autonomous Tooling

Stateful agent workflows that research, decide and act across email, WhatsApp, voice and developer environments, with an explicit operator checkpoint before any external action.

Stack

  • Python
  • FastAPI
  • LangGraph
  • Model Context Protocol (MCP)
  • Async execution
  • Human-in-the-loop
02

High-Throughput RAG & Search Engines

Retrieval, ranking and assistant layers serving real marketplace traffic in Mandarin, Spanish and English, evaluated against operator-defined quality checks.

Stack

  • Vector DBs
  • BM25 hybrid search
  • Query DSL
  • Reranking
  • LLM-as-judge evals
03

Cloud Architecture & Infrastructure Cost Optimization

Workload audits, model routing, self-hosted inference and observability that cut what AI products cost to run without cutting what they do.

Stack

  • AWS
  • GCP
  • Self-hosted vLLM / Docker inference
  • Hetzner / Docker control planes
  • Redis / PostgreSQL
  • LangSmith / Langfuse observability

Proof

04

Regulated-Domain & Commercial B2B Systems

Pricing, ERP, clinical and compliance systems for companies where the output feeds real decisions, and where reasoning, reporting and professional review stay separate and auditable.

Stack

  • ERP & pricing integrations
  • EU AI Act Article 50 transparency
  • Document intelligence designed for GDPR constraints
  • Fine-tuned VLMs (DeepSeek-VL2)
  • Human review by design
Commercial engagements

Three ways to engage.

Each engagement is scoped around a decision or a production outcome, not a headcount. Contracts are direct B2B.

A

Technical Due Diligence

Independent review of an AI product before you commit capital: is it a thin API wrapper or defensible architecture, what cloud and inference cost debt it carries, and what it would take to scale.

Format
48-hour pre-LOI sanity check · 5-day deep codebase and infrastructure audit, with written report and debrief
Best for
Private-equity deal leads, investors and acquirers evaluating an AI company or team.
Request a diligence review
B

Enterprise Production Architecture

Hands-on design and delivery of stateful agent systems, MCP server integrations and high-throughput RAG, from architecture through deployment, cost control and observability.

Format
Fixed-scope project, €800–1,500 day rate or monthly contract with retained capacity
Best for
Companies that need an AI system shipped and operated in production, not a prototype.
Scope a build
C

Fractional Technical Leadership & Advisory

Fractional CTO / CAIO ownership of architecture decisions, model and vendor strategy, AI Act readiness and delivery governance, plus executive briefings for leadership teams.

Format
€15K–€25K / month retainers · one-off executive briefings
Best for
Founders, boards and executives who need senior AI ownership without a full-time hire.
Discuss a retainer

Every engagement starts with a 30-minute scoping call. Reports and briefings are delivered in English, Spanish or Mandarin.

Book a scoping call
Featured case studies

A clinical collaboration and two production products, with the architecture left in.

Each case study covers the operating problem, the system architecture and the outcome behind it. A live WhatsApp preview fix, engineering research and an essay on multi-agent systems are on the full index.

All case studies
APOLO & medical workflow systems product interface
CollaborationTechnical lead · research and prototyping

APOLO & medical workflow systems

A family of medical-image, structured-reporting and clinical-learning prototypes with explicit human review, anchored by APOLO, a DeepSeek-VL2-tiny multimodal model for radiology and ophthalmology imaging published on Hugging Face.

The problem
Explore useful multimodal support without collapsing image interpretation, reasoning and professional review into one opaque step.
What I built
APOLO architecture and model artifacts, ophthalmology and radiology workflow experiments, reports and Umbramed learning tools.
Role: Technical lead · research and prototypingAPOLO · AutoIOL · AutoRad · UmbramedRead the case study
OULANG (欧浪AI) product interface
Personal projectProduct, engineering and operations lead

OULANG (欧浪AI)

A Mandarin-first marketplace and local-life product built for the Chinese community in Spain, live on web, iOS and Android.

The problem
Bring housing, jobs, services and community discovery into one coherent experience across languages and devices.
What I built
Web, iOS and Android surfaces with shared marketplace infrastructure, publishing, subscriptions and product analytics.
Role: Product, engineering and operations leadMarketplace · Local life · Mobile appsRead the case study
AutoClient revenue operations product interface
Personal projectProduct and engineering lead

AutoClient revenue operations

An agentic operating system for account research, enrichment, CRM work and multi-channel follow-up. A public LinkedIn recommendation cites 200+ qualified contacts in two weeks and a 30% lift in sales conversion.

The problem
Keep research, qualification and follow-up connected while preserving a clear operator checkpoint before external action.
What I built
Research and scoring workers, CRM synchronization, reporting, and email, WhatsApp and voice integrations.
Role: Product and engineering leadResearch & enrichment · CRM · Email · WhatsApp · voiceRead the case study
How I work

Staff-level AI engineering with end-to-end production ownership.

I combine agentic AI, RAG and automation with the full-stack engineering needed to put them into real products. My background in web platforms, operations and technical education supports that delivery without competing with its focus.

Hands-on by default: I move from product framing through Python and TypeScript implementation, APIs, data, deployment, observability and iteration without splitting the system into disconnected handoffs.

  1. 01

    Engineer the full AI workflow

    Retrieval, tools, structured outputs, APIs, data, interfaces, deployment and observability belong to one system.

  2. 02

    Build for operation

    Reliable AI work includes monitoring, human review and the real workflow around the model—not only a successful response.

  3. 03

    Bridge product and engineering

    Spanish, English and Mandarin delivery helps connect technical decisions with users, teams and executives.

  1. 2009–2016

    Digital design instruction · Megacursos.com

    Digital-design education supported by practical web systems and learning materials.

  2. 2016–2023

    Web platform engineering · Megawebs

    Full-stack web platforms, technical SEO, Linux operations and delivery.

  3. 2024–2026

    ML / LLMOps engineering · Agents AI Ltd

    Production AI workflows across agents, RAG, automation, deployment and observability.

Client projects, collaborations and personal products on the projects page were delivered alongside this track record as contracted, joint or owned work.

Across the public work

Follow the work where it lives.

Explore the code, technical writing, open-source tools and videos.