SELECTED PROJECTS

Proof before promises.

Each case separates the context, my contribution, the delivered result and what is not publicly measured.

01 · FINANCIAL AI AGENTS

Turning financial data into decisions.

Transform heterogeneous financial data and user documents into a structured, traceable SWOT analysis while preserving the existing analysis format and a fluid real-time experience.

My role: I develop and evolve the SVI agent for monitoring investments and holdings. I implemented automated SWOT analysis using financial data, performance, valuations, monitoring notes and user documents, with PDF and DOCX ingestion, sources, recommendations and PowerPoint export. I also integrated SSE streaming of tokens, execution steps, tool calls and SWOT results, adapted the Angular frontend, maintained a Python extraction agent for PDF, image, DOCX and PPTX files, and produced architecture documentation for the transition between Java, Python and MCP versions.
Delivered result: An end-to-end conversational workflow that enriches LLM analysis with financial and documentary context, streams its execution in real time and delivers reusable SWOT presentations.
Limits and evidence: No public metric is available; this site does not infer quantified impact.

Read the case study

SECONDARY PROJECTS

Three more proofs, three contexts.

02 · DOCUMENT INTELLIGENCE

Making documents more useful.

ContextBpifrance · PICXEL · via TALAN

My roleAs part of the PICXEL project team, I contribute to the implementation of an LLM-based document analysis solution for Bpifrance.

Delivered resultApplied generative AI experience within a business document analysis project.

View project

05 · DOCTORAL RESEARCH

Understanding performance, faster.

ContextIRISA · TARAN research team

My roleDuring my doctorate, I developed an analytical model in Python based on queueing theory. I later worked on trace windowing, Poisson-based analysis and fast trace processing as an R&D engineer.

Delivered resultThe doctoral analytical model achieved up to 500× faster evaluation than conventional simulations in the reported research comparison.

View project

03 · PERSONAL PROJECT · FULL-STACK

A smarter way to find a home.

ContextAqarih · Personal project

My roleI designed and developed the full-stack application: property listings and filters, photo management, favourites, user reviews, secure JWT authentication, Google and Apple OAuth, an administration area and an AI search assistant.

Delivered resultA production platform deployed on Vercel, with optimized image storage, transactional email and a PostgreSQL/Prisma architecture.

View project ↗

I bring a research mindset, end-to-end engineering and a focus on traceable outputs to generative AI products.

A similar context to solve?

Discuss this context