AI Engineer & AI Solutions Architect
I design and build production AI systems for organizations in Europe and the Gulf.
RAG pipelines, agentic workflows, and bilingual Arabic–English assistants — taken from requirements and architecture through deployment and evaluation.
LinkedIn mathieuajaka14@gmail.com
What I build
Four focused offers. Each links to how I build it and what an engagement looks like — if your problem doesn’t fit these, I’ll say so.
Engineering that survives production.
Not demos — deployed systems with evaluation gates, latency budgets, and data-residency constraints designed in from day one.
Bilingual retrieval, done properly
Clause-level chunking that preserves citation hierarchy, hybrid sparse + dense retrieval fused with RRF, then cross-encoder reranking — grounded generation that cites its sources and declines out-of-corpus questions.
Arabic morphology
CAMeL-based tokenization that understands roots, not just strings.
Evaluation gates in CI
Releases blocked until golden-set thresholds pass.
Latency budgets
vLLM + AWQ 4-bit on g5.xlarge, measured under load.
Model routing
Local open-weight models for routine calls, frontier APIs where they earn it.
Air-gapped & on-prem
Built for Gulf data-residency reality.
active enterprise use in 30 days
Selected work
Each case study is labeled by type, and every number shows the evaluation set behind it. No client outcomes are claimed that I can’t evidence.
Bilingual Arabic–English RAG for Saudi Labor Law
A retrieval assistant that answers labor-law questions in Arabic or English, cites the articles it relies on, and declines questions outside its corpus. Clause-level chunking, hybrid retrieval, reranking, and a bilingual evaluation set.
Agentic workflow automation for a Saudi AI agency
LangGraph-orchestrated workflows with state machines, human-in-the-loop checkpoints, model routing, and an evaluation gate on every change — built across Arabic and English pipelines.
Quantitative trading research & execution platform
A Python research and execution system for crypto perpetual futures built around leakage-resistant validation, realistic backtesting, staged promotion, and execution safety. Architecture and methodology — not a performance claim.
Every metric on this site carries its evaluation methodology — self-reported numbers say so.
How I work
The part most AI projects skip is the part between the demo and production. That’s the part I own.
-
Discovery
Understand the process today, the constraints, and what would actually be better.
-
Requirements & architecture
Scope it, then design an architecture that fits your systems and data.
-
Engineering
Build with boring, reliable tools — Python, FastAPI, PostgreSQL, containers.
-
Evaluation
Prove behavior with structured evaluation (RAGAS, LLM-as-judge, human review).
-
Deployment
Ship it into your cloud or on-prem, with the data-residency posture you need.
-
Monitoring & support
Observability, drift and cost monitoring, and human-in-the-loop controls.
At a glance
Regions
Saudi Arabia · UAE · Europe
Languages
Arabic · English · French
Core stack
Python · FastAPI · PostgreSQL · AWS · LangGraph · RAGAS
Education
MSc Applied AI · BSc AI & Business
Get in touch
For organizations
Tell me what the process looks like today, what you want to improve, and which systems or data are involved. I’ll reply with an honest read — including “you don’t need AI for this” when that’s the answer.
Discuss a projectFor recruiters & hiring managers
Hiring for AI engineering, applied AI, forward-deployed, or AI architecture work? The case studies show how I approach systems; the CV has the full timeline.
Read About & CV