At RantirAI, I was responsible for training, inference, and deployment of machine learning models for anomaly detection and natural language question answering for financial industry clients. I also developed PostgreSQL vector databases for RAG applications.
At Finsweet, I was part of the team developing wized.com and integrated AI using a custom embedding model to retrieve graph-format user workflows and process them with RAG.
At InstadeepAI, I built object-detection models for satellite imagery in the renewable energy industry and developed an LSTM demand-prediction model for solar farms. I also worked on research projects involving MCTS and deep reinforcement learning.
In my projects, I used Unsloth for LoRA fine-tuning of scripts for AI-generated code-review videos, and hand-made and coded more than 100 machine learning visualizations. I also explored weight transforms for multitask learning to address catastrophic forgetting.
