Jerome Tam

AI Product Developer – LLM-based Application

https://apps.apple.com/us/app/recipeprep/id6743941905

Solo project managed from A to Z, born from my passion for artificial intelligence and my desire to offer a concrete, useful, and accessible application for everyone. RecipePrep generates personalized recipes based on available ingredients and user nutritional goals, thanks to an embedded LLM.

-Data preparation for model training (cleaning, structuring, enrichment).

-Training of an open-source LLM with optimization techniques (fine-tuning, quantization) adapted to mobile constraints.

-Complete development of the application with SwiftUI (iOS) for the frontend, and Python/AWS for the backend.

-Deployment on AWS with a scalable, reliable, and secure infrastructure.

-Optimization of GPU usage, reducing energy consumption by 50%.

-Integration of a subscription system via StoreKit, user onboarding, nutritional calculations, and profile management.

This project allowed me to mobilize all my skills in AI, full-stack development, cloud deployment, and product design, while meeting a real user need. A complete, enriching, and entirely autonomous experience, reflecting my constant desire to learn and create with purpose.

Data & Machine Learning Analyst – Portfolio Prediction

Objective: Develop a forecasting model to optimize financial portfolio management.

Achievements:

-Collection and processing of stock market data (tech stocks: Apple, Microsoft).

-Exploratory analysis and calculation of financial indicators (return, volatility, Sharpe Ratio, VaR...).

-Creation of machine learning models (XGBoost, LSTM) to anticipate short and long-term trends.

-Optimization of model performance through feature engineering and tuning.

-Result: 20% increase in prediction accuracy through a rigorous and iterative approach.

Deliverables:

-Reusable predictive model.

-Interactive dashboard (optional).

-Detailed analysis report with recommendations.

Deep Learning Applied to Temporal Graphs — Theoretical Modeling and Evolution Prediction

Mission: Design and development of a prediction model on dynamic graph data

- Advanced research on prediction techniques based on dynamic graphs (GCN, GAT, Temporal GNN, etc.), with a focus on time-evolving structures.

- Development of an original theoretical model, integrating the temporal dimension into Graph Neural Network architectures, to capture the evolution of a graph over time.

- Targeted use case: prediction of the future state of a graph at a given time, taking into account its history and structural dynamics.

- Complete prototyping: design of a temporal data processing pipeline (graph series), training of temporal GNN models, and evaluation on simulated or real datasets.

- Result: an innovative approach combining graphs and time series, enabling the modeling of complex interactions over time and anticipating the evolutions of the represented system.