• Passionate about machine learning, time series, causal inference, and applied data analysis
• Open source projects focused on data science, modeling, and forecasting

🔗 GitHub: https://github.com/rav-lad/energy-demand-forecast

A production-grade forecasting system trained on 28,405 hourly observations (Jan 2023 – Apr 2026) of French electricity prices, demand, and weather data.

Highlights

• Designed a full MLOps pipeline covering data ingestion, feature engineering, model training, inference, and monitoring
• Implemented strict walk-forward validation over ~800 days (~19,000 prediction hours) to eliminate look-ahead bias
• Engineered features combining price lags (24h, 48h, 168h), net load, renewable penetration, weather, and calendar seasonality
• Achieved MAE = 13.41 EUR/MWh and R² = 0.831 with XGBoost, outperforming the Ridge baseline (MAE = 14.77, R² = 0.806)
• Reached 78.5% direction accuracy with Ridge Regression as a fast, robust production baseline
• Built probabilistic forecasts with LightGBM quantile regression, producing P10/P50/P90 intervals with 63% empirical coverage and 33 EUR/MWh average interval width
• Deployed a fully containerised architecture with 6 Docker services, GitHub Actions CI/CD, and a 4-page Streamlit monitoring dashboard
• Added drift monitoring and retraining alerts using PSI > 0.25 or 20% degradation in 30-day MAE

🔗 GitHub: https://github.com/rav-lad/scoring-agent

A local-first multi-agent system designed to detect abnormal and potentially informed market activity in real time across crypto markets.

The system combines statistical anomaly detection, behavioral wallet profiling, and agent-based reasoning to produce a 0–100 suspicion score, along with structured investigation outputs.

Highlights

• Built a fully local architecture powered by Ollama, ensuring privacy and low-latency inference
• Designed a multi-stage pipeline: anomaly filtering → agent investigation → deterministic scoring
• Monitors multiple data sources: on-chain flows, mempool activity, and derivatives markets
• Implements specialized agents for security checks, behavioral analysis, macro context, and quant validation
• Generates structured reports and real-time webhook alerts for high-risk events
• Includes a Rust-based terminal dashboard (TUI) for live monitoring and system observability

🔗 Orchestrator: https://github.com/rav-lad/agent-orchestrator
🔗 News Agent: https://github.com/rav-lad/news-analyst
🔗 Macro Agent: https://github.com/rav-lad/macro-analyst

A modular local-first orchestration framework that coordinates specialized AI agents to produce structured, decision-ready analysis.

The system integrates domain-specific agents (news and macro) into a unified research pipeline, enabling consistent and reproducible outputs.

Highlights

• Designed a multi-agent orchestration layer to coordinate agent execution and workflows
• Integrates specialized agents:
  • News Analyst → extracts key events, narratives, and market-relevant signals from news flows
  • Macro Analyst → provides macroeconomic context and regime-level interpretation
• Produces structured analytical outputs usable in systematic workflows
• Focus on reproducibility, modularity, and extensibility
• Fully local via Ollama, ensuring privacy and full control over models and data

🔗 Repository: https://github.com/rav-lad/Kaggle-

A collection of my Kaggle competition work, including feature engineering, model development, and evaluation pipelines across various datasets and problem types.

AWS Certified AI Practitioner

French (native), English (fluent)

Credit: Inspired by Abdul Khalid
Last Edited on: 04.04.2026