Work Experience

Key Contributions

Key Contributions

• Contributed to end-to-end data pipeline processes for a global investment/asset management client, from monthly data sourcing to client-facing dashboards delivering portfolio risk and performance insights to investors.
• Managed the monthly data provider sourcing process (medallion architecture on a custom in-house SQL Server-based platform) and oversaw provider upload compliance.
• Developed data quality controls on incoming data, and maintained/improved the SQL Server data model (tables, views, stored procedures).
• Produced and improved client-facing reporting in Excel and Power BI, combining business knowledge of portfolio risk/performance concepts with technical execution to ensure end-to-end data correctness.
• Built a pilot AI agent (Python, LangChain, LangGraph) that progressively explores a large database — schemas, then tables, then data — to answer natural-language questions from both technical users (raw queries/tables) and business users (simplified tables/graphs).
• Delivered the agent pilot successfully, demonstrating clear feasibility and value: faster build/debug cycles for the internal team and greater transparency and autonomy for the client over their own processed data.
• Built a proof-of-concept AI agent (Python, LangChain, prompt engineering) to quality-check incoming provider data, complementing existing programmatic/logical data quality checks.
• Diagnosed and automated numerous manual, tedious business-as-usual processes throughout the engagement using Python scripts, integrated into the main application or delivered as standalone tools.

Key Contributions

• Built a client satisfaction tracking web application (Angular, Spring Boot, PostgreSQL, Leaflet) enabling staff to log client interaction outcomes (satisfaction level + free-text notes) after every client touchpoint.
• Designed a live, always-on map of client satisfaction across France (metropolitan + overseas territories), selectable by time period, displayed company-wide so staff could check client sentiment before meetings and calls.
• Integrated a lightweight, locally-run Hugging Face NLP model (keyword extraction + sentiment analysis) to automatically process free-text notes into word clouds, surfacing recurring themes without manual review.
• Outcome: improved internal communication and client relationship tracking across teams; delivered a complete, demo-ready application by end of contract, laying the groundwork for a planned production rollout.

Key Contributions

• Main mission: built an interactive dependency graph web application (Angular, Spring Boot, Oracle DB, Sigma.js, Graphology) visualizing software release dependencies as a directed graph, replacing a fully manual, error-prone process based on non-standardized PDF delivery notes.
• Developed a Java regex-based parser to extract release version and dependency data from inconsistent PDF delivery notes, and drove alignment across development teams on a standardized delivery note format.
• Built graph app features: fuzzy/typo-tolerant search (edit-distance algorithm), interactive zoom/pan with dependency highlighting, timestamped and undoable change history, color-coded node categorization, an interactive statistics dashboard, and multiple graph layout algorithms (ForceAtlas2, Dagre).
• Delivered iteratively using Agile methodology, presenting and validating each feature with the team; used Balsamiq for mockups and Git for version control.
• Outcome: gave installation technicians a reliable, self-service way to check release dependencies, speeding up and simplifying installations while reducing the risk of installing incorrect dependencies.
• Secondary mission: worked alongside an architect on a proof-of-concept to restructure the release packaging pipeline, migrating toward Maven + RPM/Yum-based dependency resolution for Linux package installation.
• Organized and led cross-team alignment meetings to define and get adopted a standardized XML schema for release/dependency metadata; authored the structural proposal that was ultimately adopted.
• Built a custom Maven plugin that consumes the standardized XML metadata to auto-generate uniform PDF delivery notes and insert release/dependency data directly into the database, eliminating manual PDF parsing for releases using the new pipeline.
• Self-initiated an AI innovation project: tested zero-shot prompting with ChatGPT for dependency extraction, found it unreliable, then designed and executed a supervised fine-tuning pipeline for an open-source LLM (Llama 2, 7B, via Hugging Face) on Google Colab, using a dataset auto-generated from the existing regex parser.
• Result: the fine-tuned 7B model reliably outperformed general-purpose 175B-parameter models on the narrow extraction task, demonstrating the value of task-specific fine-tuning; presented the POC company-wide to promote practical AI adoption.
• Built two standalone JavaScript/Node.js automation tools to speed up colleagues' recurring manual file-processing tasks, packaged as cross-platform executables (Windows/Linux/macOS).

Key Contributions

• Improved a WPF desktop application (C#, XAML) connecting to highway systems for real-time monitoring and control: redesigned the UI/UX with mockups validated by the team before implementation, applying Material Design principles.
• Cleaned up the existing codebase, added new features, and fixed bugs in the monitoring application, resulting in improved UX and faster day-to-day use for technicians.
• Applied the same redesign, cleanup, feature, and bug-fix process to a second internal tool used to encrypt SQLite databases (C#, XAML, WPF, Material Design).
• Outcome: faster and easier monitoring workflows for technicians, contributing to faster client support and intervention on highway systems.