About

I'm looking to join a team that designs AI agents and ships them to production: conversational assistants, RAG, multi-agent systems, connected to the tools people use at work. I'm here to learn: on a real codebase, with code reviews, production constraints and people more experienced than me. On a work-study contract (French contrat de professionnalisation) to complete my training at Ada Tech School, or on a permanent contract. For ten years in asset management, I translated complex strategies for clients; today I do the same work between a business need and the agent that answers it, with verifiable sources and tests. I have built mostly on AWS, and I'm ready to learn the team's stack: another cloud, another framework, another language.

Role

AI Developer · AI Engineer: agents, RAG, multi‑agent systems, AI built into applications

Sectors

Open to any sector; an edge in finance, insurance and regulation (ten years in asset management)

Stack

Mostly AWS (Bedrock, AgentCore, Strands) · Claude · Ollama for open-source models · MCP · pytest and GitHub Actions; ready to learn Azure, GCP and the team's tools

Contract

Work-study contract of 12 to 24 months (Ada Tech School training), or permanent contract

Schedule

4 days at the company, 1 day at school

Where

Paris area, hybrid; happy to travel for work

Availability

Immediately

Languages

French (native) · English, fluent, my working language · Mandarin Chinese (beginner)

• Building AI agents

  In my Nanodegree, NovaMart needed customer support that could handle orders, returns and questions about its terms. My task: a multi-agent system on AWS that only answers from verified facts and resists hijacking attempts. I built five agents coordinated by an orchestrator, three document bases queried in parallel, a guardrail against banned topics and personal data, and tracing of every step in X-Ray. Result: 120/120 on the project tests, 14/14 on the guardrail tests, and a project approved by the Udacity mentor.

  AWS 3

• Explaining simply

  In asset management, I already translated the strategies of portfolio managers and quants for clients. As I trained in AI agents, concepts piled up fast, each tied to a tool or a cloud. I set myself a goal: keep only what holds for any agent, and explain it simply. I wrote BAA 101: each chapter starts from a real project, stripped of its specifics, and a 54-term glossary opens as a bubble in the text. Result: eleven chapters published, in French and English.

  BAA 101Glossary

• Rigour

  During Dev3Pack, my agent ran on a model installed on my own machine. I had to set a time limit after which it gives up, or it could wait forever. I had an estimate in mind; rather than trust it, I measured: 1.9 seconds on the first call, 0.3 after that. Result: a 3-second limit based on measurement, while my estimate had been wrong. Same reflex when a test hadn't run: I declared it instead of making up a number. On my Gecko buyer agent, 98 tests run on every push in GitHub Actions. And when an LLM judge scored my support chatbot 0.92, I still checked the database: it had let fake ticket IDs through.

  Dev3Pack S2S8Gecko · CI ↗Chatbot · LLM judge ↗

• Team spirit

  At the selection hackathon for the Dev3Pack accelerator, we were building SafeHaven Money when my teammate's computer gave out. We still had to deliver the app on time. I took over all the work and finished it with Codex. Result: the app was delivered on time and presented to the jury, and I joined the Dev3Pack accelerator. During the Nanodegree, I also answered other students' questions on the Udacity forum.

• Understanding the business before coding

  A business school degree with a Finance major, an MBA in Trading & Asset Management, then ten years in asset management, including at Crédit Agricole CIB, Candriam and Northern Trust. My job was to understand what a client needed, then explain it to them without jargon. That is where a useful agent starts: knowing which question the business is asking before choosing a model. I know institutional finance and its regulation in particular, and I also build on blockchain: an app on the XRP Ledger, a contract deployed on Ethereum (Sepolia testnet), and an agent that buys on-chain on Solana.

  B-Wiize ↗SplitBook ↗Gecko

• Innovation

  In late July 2026, in the middle of my Nanodegree, AWS put the tool my first project was planned on into maintenance. I still had to deliver a support chatbot that sends each message to the right behaviour. Rather than patch the old approach, I rebuilt the project on AgentCore's managed harness, then brand new: all the routing lives in the prompt's rules. Result: an approved project, scored 0.92 in automated evaluation.

  AWS 1

• Giving back

  Women In DeFi was looking for mentors for its Intro to Web3 Squad 4.0 programme. I applied and led a 90-minute session for a group of about 400 participants: blockchain basics, then confidence in career choices and visibility. Women In DeFi gave me a certificate of appreciation. During the Dev3Pack bootcamp, I also reported two course bugs in public issues, with their cause and a proposed fix. And I write on Dev.to about what I learn while building.

  Women In DeFiIssue #15 ↗Issue #16 ↗Issue #12 (Gecko) ↗Dev.to ↗

1. 2012 – 2016

   Starting out in finance

   Master in Management (Grande École programme), Finance major, at Clermont School of Business (formerly ESC Clermont) in 2015, then an MBA in Trading & Asset Management at ESLSCA Business School Paris in 2016. First roles in Paris, at Crédit Agricole CIB, then in Tokyo, in financial advisory.

2. 2017 – 2020

   Luxembourg

   Junior Client Portfolio Manager on multi-asset funds at Candriam, then institutional sales at Northern Trust.

3. 2019 – 2021

   My first venture

   On the side, Mialy Ericka: digital communication strategy for women entrepreneurs.

4. 2022 – 2023

   Paris

   Wealth management advisor.

5. 2024

   I switch to code

   Fullstack developer training at Ada Tech School. The diploma is still to be completed, through a work-study contract.

6. 2025

   Discovering Web3

   This is where I discover blockchain and the world of Web3: the Aquarium residency at XRPL Commons, then founding B-Wiize.

7. 2026

   From Web3 to AI agents

   I join the Dev3Pack ecosystem in January, with a first bootcamp to learn Solidity, then follow SheFi (season 15), a DeFi education programme. Then come SafeHaven Money, a mentoring session at Women In DeFi, AWS AI & ML Scholars and, on my own initiative, Dev3Pack's AI Engineering bootcamp.

• since September 2026

  Dev3Pack

  Country Lead, France chapter

  I have known the Dev3Pack ecosystem from the inside since January 2026. Today I run its French community: I recruit participants and organise events. It's something I do alongside my work, because I care about it.

• since May 2026on pause

  SafeHaven Money

  co-founder and CEO

  A voice-first stablecoin platform for freelancers and entrepreneurs in countries where inflation eats into their income. Built on Solana, with LI.FI.

  Code ↗
• May 2026

  Women In DeFi

  mentor, Intro to Web3 Squad 4.0

  A 90-minute session for a group of about 400 participants: blockchain basics, confidence in career choices, and visibility.

• 2025

  B-Wiize → Pop Wallet

  founder

  A gamified Web3 app that teaches students to manage their budget, built on the XRP Ledger. In late 2025 it became Pop Wallet, with its first ten users.

  B-Wiize code ↗

• Nick Saraev

  YouTube ↗

  For his clear explanations of how AI agents work inside. He makes a technical point easy to follow, and even interesting.

• Marina Mogilko

  Silicon Valley Girl ↗

  I've followed her for seven years. I value how she has grown, her insights and her talks from inside Silicon Valley. She is the one who confirmed my wish to move into tech.

Each tool links to where I used it.

• Udacity · AWS

  AWS AI & ML Scholars

  A two-stage programme: a challenge open to all, then a Nanodegree for the top performers.

  The programme on AWS's website ↗
  1. 
     Stage 1 · Course certificate · June 13, 2026AWS AI Practitioner Challenge

     From March to June 2026, 100,000 of us took on the challenge, with two projects to complete.

     Verify ↗
  2. 
     Stage 2 · Nanodegree · October 6, 2026Future AWS Agent Engineer

     The top 4,500 from the challenge were selected for this Nanodegree. I saw it through: three projects approved by the Udacity mentor.

     Verify ↗
  The book of this programme →
• Dev3Pack

  AI Engineering Bootcamp

  1. 
     Certificate of completion · October 2026

     Three weeks (September 14 → October 2, 2026) to build a source-grounded research agent. Graded track 5,100/5,100, final assignment 15/15, Gecko capstone delivered and defended orally.

     ID 0MEIAR4GV9KFVerify ↗15/15 result ↗
  The book of this programme →
• Cyfrin

  Cyfrin Updraft

  Three blockchain development courses. Each badge certifies that the course was completed and the proficiency exam passed.

  1. 
     Stage 1 · Completion badge · February 5, 2026Blockchain Basics

     Course completed, proficiency exam passed.

     ID WFISK3J5PLXZVerify ↗
  2. 
     Stage 2 · Completion badge · February 15, 2026Solidity Smart Contract Development

     Course completed, proficiency exam passed.

     ID KDHNY1XK0GNHVerify ↗
  3. 
     Stage 3 · Completion badge · March 1, 2026Foundry Fundamentals

     Course completed, proficiency exam passed.

     ID B4K2I50QC0TIVerify ↗
• SheFi

  SheFi Season 15

  1. 
     Certificate of completion · February 2, 2026

     A DeFi and Web3 education programme, taken early in the year, before moving on to AI agents.