CyberShen2024–2026
SAP
Full-Stack Software Engineer
Built and shipped product features in an international environment, while evolving an established architecture safely.
Java · Spring Boot · Kafka · TypeScript · AngularAvailable for full-time roles
Ilian Allaf · Full-stack software engineer
I build products that hold up in production.
From interface to infrastructure, I turn complex problems into clear, reliable software.
France
Products
Experience
Selected work
Founder & product engineer.
End-to-end ownership, from idea to production.
01Photography product
Photorunning
A multi-tenant photography platform capable of ingesting up to 400,000 images per event and retrieving them by race bib or face.
Next.js · tRPC · AWS S3 · SQS · Lambda · Rekognition
View case study
02AI accounting platform
Soomeo
An accounting platform that transforms uploaded invoices into structured, validated, and actionable financial data.
Next.js · tRPC · PostgreSQL · Prisma · AWS S3 · Lambda · CloudWatch
View case studyExperience
Software engineer.
Shipping within established teams, systems and constraints.
2023
CyberShen
Full-Stack Software Engineer · Internship
Built a Microsoft 365 security scanner, from its data model and detection rules to the review interface.
Golang · GraphQL · SQL · Next.js · KubernetesMy AI practice
How I work with AI.
A personal practice shaped by curiosity, judgment and experimentation.
I experiment before I standardise.
I enjoy trying new ways of building with AI rather than committing too early to one method. I have spent time with BMAD, spec-driven development, agentic workflows and multi-agent coordination. Some ideas become part of my practice; others teach me where process starts adding more ceremony than value.
I keep learning from practitioners.
The field changes too quickly for a fixed playbook. I follow engineers sharing real work on X, read technical blog posts and watch hands-on YouTube breakdowns. I am most interested in what survives contact with a real codebase, not just a polished demo.
What I believe
AI works best when the engineering underneath is already sound.
In my experience, AI accelerates good engineering far more than it compensates for weak foundations. I rely on SOLID to keep responsibilities and dependencies explicit, DDD to keep business language and rules coherent, and Clean or Hexagonal Architecture to protect the domain from infrastructure details. Those boundaries help the agent understand the system while helping me review its work.
- SOLID
- DDD
- Clean Architecture
- Hexagonal Architecture
Why tests matter with AI
A good test setup makes AI more useful.
I see tests as the feedback loop that makes AI-assisted development trustworthy. When the important behaviours and boundaries are already covered, an agent can iterate faster, detect regressions earlier and work against clear acceptance criteria. This is also why I value TDD: defining behaviour before implementation gives both me and the agent a tighter target and a faster feedback loop. The goal is not coverage for its own sake, but relevant feedback on the risks of each change.
- Unit
Domain rules and edge cases
- Integration
Databases, queues and providers
- Contract
APIs and event boundaries
- End-to-end
Critical user journeys
- Non-functional
Performance, resilience and security
The workflow I keep coming back to
I like Matt Pocock's workflow because it keeps the thinking visible.
I do not run every step mechanically. For a small change, that would be unnecessary. But when work spans several sessions, I like how the workflow separates deciding from building, preserves the reasoning, and keeps each implementation step small enough for me to inspect properly.
- 01
/grill-with-docsClarify the problem, vocabulary and hard decisions.
- 02
/to-specTurn settled decisions into durable context.
- 03
/to-ticketsSplit the work into testable vertical slices.
- 04
/implementBuild one bounded outcome at a time.
- 05
/code-reviewReview the diff against the original intent.
Contact