Skip to content
Yassir Halaoui
FR

Yassir Halaoui / Senior Java & React developerBuild the system.Ship the product.

Secure systems for banks and payments. AI agents, technical SEO, and products of my own, built and launched end to end.

Explore my workHire me in Paris
Paris, France9 years in production · AWS Certified

Three products, built and run alone

Each one designed, coded, shipped and operated by me: payments, email, SEO and support included. They're the best evidence I have of how I work when nobody else is in the room.

macOS app · on-device AI

Kempt

Your Mac's paperwork, named, found and ready to send. Apple's on-device model reads every document, so nothing ever leaves the machine.

  • Runs entirely on your Mac, no account
  • Every answer cites the page it came from
  • 122 automated tests

Read the case studyVisit usekempt.app

  • Kempt's Dossiers view: the documents a landlord asked for, found on the Mac and ready to export as one dossier.
  • A pay stub saved from an email: Kempt names it in the Mac's notch and files it in Work.
  • Asking Kempt where the lease is: the file, its rent and dates, and the sentence the answer came from.
  • A messy Downloads folder dropped on Kempt: four files named and filed in one pass.
  • A Mac desktop with a payroll email open: the attached pay stub is about to be saved, and Kempt will name and file it.

  • Dossier

    Ten documents by tonight. Kempt reads the request and builds one PDF.

  • Ask

    What's my rent? The answer, with the page it came from.

  • Names

    A wall of ugly file names, renamed by what's inside them.

  • Zero

    Tax notice, payslips, passport. Bytes sent anywhere: zero.

How Kempt works

  1. The AI lives on your Mac

    Kempt uses the Apple Intelligence model built into macOS. Your payslips, IDs and bank details are read right there on the machine. Nothing is uploaded, there's no account, and there's no cloud bill behind it.

  2. Fast enough to feel instant

    Ask a question and the first answer arrives in 25 milliseconds. A full clean-up scan of the Mac takes under a second and uses 83 MB of memory, down from 8.2 seconds and 1.7 GB in the first build.

  3. Smart where it helps, strict where it counts

    The AI writes the file names. The facts on the page (dates, amounts, who it belongs to) decide where a file goes. Every answer shows the document it came from, and anything Kempt isn't sure about waits for you.

First answer
25 ms
Full clean-up scan
0.86 s
Files uploaded
0
Memory for a full scan
83 MB

Real-time web app · 3 languages

RevealTogether

One secret, every screen, the same second. A live gender reveal where every guest votes, counts down and finds out together, wherever they are.

  • 1,000+ families in 40+ countries
  • English, French and Spanish
  • Page one on Google in English and Spanish

Read the case studyVisit revealtogether.com

  • RevealTogether's homepage: One secret. Every screen. The same exact second.
  • RevealTogether's pricing: four one-time plans, from 7.99 dollars to the Family Pass.
  • RevealTogether's free planning tools: due date calculator, pregnancy countdown, boy or girl poll, gender predictor.
  • The live countdown at 3, guests split between Team Boy and Team Girl with their votes counted.

B2B SaaS · AI assessments

TheQuizMaster

A job description in, a proctored technical assessment out, graded against a rubric anyone can read. Candidate code actually runs.

  • Candidate code executed in 14 languages
  • An integrity score, never a verdict
  • Model calls run in parallel

Read the case studyVisit thequizmaster.io

  • TheQuizMaster scoring a candidate's intro video: communication 80 out of 100, broken down into fluency, vocabulary, structure and clarity.
  • A full-session integrity audit: skill 84 times integrity 0.72 gives a final score of 60, with the timeline of flagged moments.
  • Four steps: paste the job description, get a generated assessment, invite candidates, read the insights.
Film · Why we built TheQuizMaster1:24
Yassir Halaoui waving while introducing TheQuizMasterWhy we built TheQuizMasterWatch on Vimeo
Yassir Halaoui explains why TheQuizMaster exists: technical hiring leans on take-home tests nobody trusts, so the product generates the assessment from the job description, runs the candidate's real code and scores it against a rubric you can read.

What you can hire me for

Four ways to work together, each one backed by something already shipped.

  1. Java & React in ParisNine years of Java 21, Spring Boot, Kafka and React in banks, payments and insurance. On-site in Île-de-France or remote.9 yearsJava and React in production5Employers and consulting missionsJava 21 · React 19Current stack, in production
  2. AI agents & LLMsAgents and LLM features that cite their sources, run on your data and get measured before they ship.KemptOn-device AI. Answers cite their source.TheQuizMasterLLM-built assessments in 14 languages4 daysKempt, built with AI coding agents
  3. Technical SEORendering, structured data, hreflang and AI-search visibility, fixed in your codebase by the person who'd write it.Page oneRevealTogether, in English and Spanish350+Articles published in three languages6 weeksNew domain to page one for my name
  4. Zero to MVPFrom idea to paying customers, end to end. Kempt took four days from idea to launch.4 daysKempt, from idea to launch1,000+Families on RevealTogether, 40+ countries3Products built and run alone

Competences

Every tool on my CV, grouped by where it has actually been used in production.

Every competence
  1. 24 toolsJava & Spring BootJava 8 to 21, Spring Boot 3.5, Spring Batch, JPA and AOP, in banking and payments since 2017.
  2. 27 toolsReact, Next.js & TypeScriptReact 17 to 19, Next.js, strict TypeScript and Module Federation micro-frontends.
  3. 14 toolsKafka & event-drivenKafka, Debezium CDC, the transactional outbox and Avro schemas, built for payment platforms.
  4. 22 toolsAPI security & integrationOAuth 2.0, OIDC, RBAC, mTLS, WS-Security over SOAP, API gateways and webhooks.
  5. 28 toolsCloud, Kubernetes & DevOpsAzure AKS, AWS, GCP, Kubernetes, Helm, Terraform and CI/CD with quality and security gates.
  6. 15 toolsPostgreSQL & dataPostgreSQL in depth (JSONB, WAL, replication slots), plus Oracle, Redis and Elasticsearch.
  7. 14 toolsSwift & native macOSNative Mac apps in Swift and SwiftUI with Apple's on-device model, notarized and shipped.
  8. 12 toolsAI agents & LLMsAgents and LLM features that cite their sources, run on your data and get measured before they ship.
  9. 15 toolsTechnical SEORendering, structured data, hreflang and AI-search visibility, fixed in your codebase by the person who'd write it.

Nine years where failure was expensive

Two banks, a payments group, an insurance platform and a consulting mission. All regulated, all audited.

  1. Aug 2025 — Present

    Pluxee

    Senior Fullstack Java / React Developer

  2. Jul 2024 — Jul 2025

    Docaposte

    Fullstack Java / React Developer

  3. Aug 2022 — Jun 2024

    Société Générale

    Lead Developer, then Fullstack Java / React Developer

  4. Feb 2022 — Aug 2022

    Talentum Technology Services

    Senior Java Consultant

  5. Jul 2017 — Jan 2022

    CIH Bank

    Software Engineer

Full CV

Field notes

What broke in production, and why. Notes, not tutorials.

All field notes