Mouhssine Aregu

• Engineered High-Fidelity Workspace: Designed a production-grade platform for autonomous agent design. Tabbed interface (Config/Training/Sandbox) with reactive state synchronization and dynamic capability injection (MCP, Skills, Knowledge).

• Cognitive Memory & Context Intelligence: Implemented a 'Persistent Brain' (long-term memory) and a contextual compaction engine (automatic summarization at 75%+ tokens) ensuring infinite coherence on large-scale projects.

• LLM Fine-Tuning & Alignment (SFT/LoRA/DPO): Refined models via supervised fine-tuning pipelines (SFT/LoRA) and preference optimization (DPO) for precise alignment of agents to business standards.

• Hybrid RAG & Knowledge Graph (Deep Analysis): Cutting-edge pipeline merging Vector Search (Qdrant) and Graph Database. AST parsing via Tree-sitter and re-ranking by Cross-Encoders for surgical precision on massive codebases.

• Expert Agent Fleet & Audit Engine: Orchestrated a specialized fleet (Skill Specialist, Workflow Architect, Documenter, Critic) automating technical audits and strategic decision support via AI.

• Multi-Provider Reliability Engine: Agnostic architecture supporting market leaders (OpenAI, Claude, Gemini) and Local-LLM (Open Source). Includes a Universal Transformer (XML/JSON normalization) and encrypted secret management.

• Clean Architecture & Scalability: Robust backend using FastAPI/SQLModel, async orchestration via Celery, and global diagnostics for maximum application reliability.

- Languages & Tools:

Python, FastAPI, SQLModel, Pydantic, Alembic, Tree-sitter, Uvicorn, Celery, Psutil.

ReactJs, Next.js, Vite, Tailwind CSS, Lucide Icons, Recharts, Prism, WebSockets.

Ollama, langGraph, Qdrant, FalkorDB, Sentence-Transformers, Cross-Encoders, SFT/DPO, LoRA.

Stripe, AWS, Lambda, Git Flow, Docker, Node.js, Npm, VS Code, Agile (Scrum).

- OS: macOS / Linux

Project:

ART: Asset Rotation Tools supports all the Financing Bank's business lines by distributing bank financing and intervening in both the primary and secondary markets, through cash and synthetic solutions.

Contribution:

- Industrialization of an algorithm based on Machine Learning and Operational Research on Databricks.

- Integration of an LLM to automate financial audits and improve decision support.

- Implementation of Celery to orchestrate asynchronous tasks and optimize background processing.

- Complete management of the product development cycle, from design to production.

- Development of high-value-added features, adapted to the specific needs of the product and users.

- Advanced data manipulation: expertise in processing and transforming large-scale data.

- Design of a notification system with automatic email sending.

- Development of a tool for automatic generation of Excel reports.

- Creation of Python scripts for mass data processing.

- Implementation of a Data Lake integrating multiple sources and databases.

- Integration of an SSO authentication system.

Project:

Odyssey: an ambitious project within the IT Resort department of Club Med's IT. Its main objective is to modernize the currently used PMS (Property Management System), which runs on the Magic language, to a more modern and performant web technology. Currently, the PMS is used in all Club Med villages to manage their entire activities and simplify operations to enhance customer experience.

Contribution:

- Take charge of the entire product development cycle, from the design phase to production.

- Actively participate in code reviews, ensuring the overall quality and stability of the product.

- Develop high-value-added features, meeting the specific needs of the product and users.

- Ensure the complete realization of developed features, overseeing their successful deployment to production.

- Continuously improve code performance and overall product design, implementing best development practices.

- Fully participate in the creation of end-to-end (E2E) functional tests with Cypress.

- Implementation of an error tracking system with Sentry.